Having spent last week in London for a series of talks that Jefferies invited me to give to investment professionals, I'm struck by how quickly the conversation around AI in investment management has matured. Six months ago in Hong Kong, the focus was on mapping the future, defining the horizons of what AI could do. Now, the questions are far more urgent and pragmatic: Why are so many firms, despite significant investment, struggling to see a real return? And what separates the handful of teams achieving breakthrough results from the vast majority stuck in pilot mode?

The answer is that we’ve entered a new phase. The industry is splitting, and a great "GenAI Divide" is opening up between those who are merely augmenting old processes and those who are fundamentally transforming them. A recent MIT report brought this into sharp focus with a sobering statistic: 95% of organizations report getting zero measurable ROI from their GenAI investments.

This isn't an indictment of the technology. It's a reflection of a profound challenge: How do you hit a target that’s not just moving, but accelerating?

The Paradox: Why Progress is Becoming a Problem

I've lived through the PC, Internet, and Mobile revolutions, but the defining characteristic of this new AI era is its breathtaking velocity. New models and capabilities are released weekly, not yearly. While exciting, this relentless pace creates a paradox. For many organizations, the sheer speed of innovation has become a barrier to adoption. It’s nearly impossible to build a stable foundation of skills when the ground is constantly shifting beneath your feet.

This leads to what I see as three critical barriers:

1. A Technology Barrier (The "Learning Gap"): Most enterprise tools are static; they don't learn from user feedback. This is why 35% of employees are paying for their own AI tools—the ones their companies provide aren't good enough.

2. A Human Barrier (The "Proficiency Gap"): We are deploying revolutionary technology to a workforce that hasn't been adequately trained. It’s not about knowing which button to click; it’s about understanding how to structure a problem and collaborate with the AI.

3. An Organizational Barrier (Silos & Friction): Without a top-down mandate, AI initiatives often become disjointed science projects that fail to deliver a cohesive, firm-wide advantage.

The temptation is to wait for things to "settle down." But that’s precisely the wrong strategy. We have to accept a new reality: the AI you use today is the worst AI you will ever use. The window for competitive advantage is closing, and the greatest danger is standing still.

The "Swiss Army Knife" Strategy: A Playbook for Today

So, how do you move forward? You don't need to wait for the perfect, purpose-built "AI for Asset Management" platform to magically appear. The path forward lies in what I call the "Swiss Army Knife" strategy.

In the early PC era, the spreadsheet was the original Swiss Army Knife. It was a powerful, general-purpose tool we used to build incredibly complex applications because the specialized software hadn't been invented yet. Today, the major LLM platforms—Claude, Gemini, ChatGPT—are the new Swiss Army Knives. They are immensely capable and can be adapted to solve complex problems right now.

We put this to the test at 71 Capital. We chose one of the most complex, time-consuming tasks we face: synthesizing a chaotic virtual data room into a 50-page, institutional-quality investment memo.

The data room was a mess of PDFs, spreadsheets, and transcripts totaling over 21 million tokens—far too large for any model's context window. This is where we discovered the single most important skill in this new paradigm is not "prompt engineering," but "Context Management."

You can't fit the ocean in a bucket. The real, high-value human skill is intelligently selecting the right combination of documents—the financials, transcripts, and reports most relevant for each specific section of the memo. It’s about being a brilliant editor and curator of information before the AI even sees it.

By breaking the problem down section-by-section and feeding the model the precise context it needed for each part, we transformed a multi-week, multi-person process into something that could be largely completed in a day. This wasn't a 20% productivity boost; it was a 10x force multiplier.

From Manual Workflows to an "AI Deal Room OS"

Our experiment was powerful, but still manual. It required a lot of copying, pasting, and managing versions. But it validated a vision for the next logical step: an "AI Deal Room Operating System." Imagine a single, end-to-end system that ingests a messy data room and, guided by a human analyst, uses a team of specialist AI agents to draft a fully cited, institutional-quality memo.

This isn't science fiction. The most forward-thinking firms are already building toward this, moving beyond simply sprinkling AI on legacy platforms and instead, architecting new, AI-native workflows from the ground up.

During the panel discussion at Jefferies, this raised a critical question about the future role of the analyst. Does this automation eliminate the need for junior talent? I believe the opposite is true. By automating the "grunt work" of data collation, we liberate analysts to focus on higher-value tasks: deeper critical thinking, meeting with management teams, and—as one panelist astutely noted—perhaps even getting out of the office to find physical insights you can't get from a screen. The job becomes less about finding the data and more about what you do with it.

Will You Cross the Divide?

The investment industry is at a strategic inflection point. The choice is no longer if you will adopt AI, but how. Will you stay on the side of incremental augmentation, getting marginal gains from legacy tools? Or will you cross the divide and embrace a transformative approach?

The path to a 10x advantage requires a shift in mindset. It means empowering your teams with new skills like Context Management. It means reimagining core workflows. And it means starting now.

I’ll leave you with a thought from my former mentor at Intel, Andy Grove. He used to say that in times of great change, "Your tendency will almost always be to wait too long. Yet the consequences of being early are less onerous than the consequences of being late."

With AI, the greatest danger truly is standing still.

More critically, we are merely in the "1960s" of this new computing era—a sobering reality check for those racing to build or fund the next AI unicorn.

The Strategic Inflection Point: From Code to Weights to Prompts

Karpathy's framework elegantly captures three distinct programming paradigms that now coexist in our technological landscape:

Software 1.0: The Classical Era Traditional programming where humans write explicit instructions in languages like Python or C++. This is the GitHub universe—millions of repositories containing human-crafted logic.

Software 2.0: The Neural Network Revolution Here, the "code" consists of neural network weights, not written but optimized through data and training. Hugging Face has emerged as the GitHub equivalent for this paradigm. At Tesla, Karpathy witnessed this transformation firsthand as neural networks literally "ate through" the autopilot's C++ codebase, replacing traditional algorithms with learned representations.

Software 3.0: The Natural Language Disruption The game-changer: Large Language Models programmable through natural language prompts. For the first time in computing history, English has become a programming language. This isn't merely an incremental improvement—it's a fundamental reimagining of human-computer interaction.

Strategic Implications for Product Development

The emergence of Software 3.0 creates unprecedented opportunities and challenges:

• Democratization of Development: "Vibe coding" enables non-programmers to build functional applications • Paradigm Fluency: Successful products will seamlessly blend all three paradigms • Infrastructure Transformation: Existing digital infrastructure requires fundamental adaptation

Understanding LLMs: Beyond the Utility Metaphor

Many investors and builders make the critical error of viewing LLMs as simple utilities—commoditized services differentiated only by price and performance. Karpathy's analysis reveals a more nuanced reality.

The Operating System Paradigm

LLMs are evolving into complex software ecosystems that mirror the structure of operating systems:

Architectural Parallels:

• LLM core = CPU equivalent

• Context windows = Working memory

• Tool integration = System calls and APIs

• Multi-modal capabilities = Device drivers

Market Structure Implications:

• Consolidation around few major providers (OpenAI, Anthropic, Google)

• Open-source alternatives emerging (Llama ecosystem as the "Linux" of LLMs)

• Application layer opportunities expanding rapidly

This framing has profound implications for investment strategy. Just as the OS wars of the 1980s-90s created massive value in the application layer, we're entering a similar phase with LLM platforms.

The Partial Autonomy Imperative: Building Products That Actually Work

Perhaps Karpathy's most actionable insight concerns product design philosophy. Drawing from his five years building Tesla's autopilot, he articulates why "partial autonomy" represents the optimal approach for current AI products.

Case Study: The Anatomy of Successful LLM Applications

Examining leaders like Cursor (coding) and Perplexity (search) reveals consistent design patterns that separate winners from also-rans in the AI application space.

The Email Management Paradox: A Tale of Two Approaches

Consider the contrasting philosophies of two AI email assistants: Jace and Cora. Their divergent approaches illuminate why Karpathy's principles matter for product success.

Jace: The Cursor of Email Jace explicitly models itself after Cursor's successful design patterns:

• Three-pane interface: Email categorization (left), content and editable draft (center), chat-based instructions (right)

• Real-time verification loop: Users see and modify AI-generated responses before sending

• Multi-tool orchestration: Integrates calendar analysis, web search, and file processing

• Contextual learning: Adapts to individual communication styles while maintaining human oversight

Users report saving hours per week on email with minimal editing required—the hallmark of effective human-AI collaboration.

Cora: The Scheduled Summary Approach Cora takes a fundamentally different path:

• Passive consumption model: Delivers twice-daily email summaries

• No real-time interaction: Users receive briefings at 8 AM and 3 PM

• Limited verification opportunities: Drafted responses exist but lack immediate editing interface

• Philosophy-driven: Forces users into a specific email methodology

The market response reveals which approach resonates. While Cora frames itself as "not an 'AI email assistant'" but rather an "email methodology," Jace users describe it as transformative for actual productivity. The difference? Jace enables the rapid generation-verification loop that Karpathy identifies as critical.

Universal Success Patterns

1. Context Management Sophistication The LLM handles complex context automatically, abstracting away the complexity from users while maintaining transparency.

2. Multi-Model Orchestration Successful apps don't rely on a single LLM but orchestrate multiple models for specialized tasks—embeddings for search, generation for drafting, analysis for categorization.

3. Purpose-Built Interfaces Generic chat interfaces are insufficient. Cursor's diff visualization, Perplexity's source citations, and Jace's three-pane layout exemplify how domain-specific UIs enable rapid human verification.

4. The Autonomy Slider Users control the level of AI involvement, from minor assistance to full automation. This isn't a compromise—it's a feature that acknowledges the current reality of AI capabilities while building toward greater autonomy.

The Verification Bottleneck

Karpathy's key insight: the human verification loop is the critical constraint in AI-assisted workflows. Products must optimize for:

• Rapid visual assessment (leveraging human "GPU" for pattern recognition)

• Incremental, auditable changes (avoiding overwhelming diffs)

• Failure-tolerant workflows (assuming AI errors as default)

The Jace-Cora dichotomy illustrates this perfectly. Cora's twice-daily briefings break the verification loop, forcing users to context-switch hours after emails arrive. Jace, by contrast, enables immediate verification and editing—maintaining the flow state that defines productive work.

This isn't merely a UX preference; it's a fundamental requirement for human-AI collaboration. As Karpathy observed from Tesla's autopilot development, the path to full autonomy requires thousands of iterations of the human-AI loop. Products that optimize this loop will compound their advantages over time.

The Infrastructure Revolution: Building for a Multi-Agent Future

Forward-thinking organizations must recognize that agents represent a new class of digital infrastructure consumers, distinct from both humans (who use GUIs) and traditional computers (which use APIs).

Strategic Infrastructure Adaptations

Documentation Revolution:

• Markdown-first documentation (LLM-readable)

• Replacement of "click here" instructions with executable commands

• Early movers like Vercel and Stripe gaining competitive advantage

New Protocols and Standards:

• llm.txt files for domain-specific LLM instructions

• Anthropic's Model Context Protocol for agent communication

• Ingestion tools that transform human-readable content to LLM-optimized formats

The Competitive Imperative: Organizations that fail to adapt their digital infrastructure for agent consumption will find themselves at a severe disadvantage as AI-assisted workflows become standard.

Investment Thesis: The Decade of Agents

Karpathy's framework suggests several key investment principles:

1. Realistic Timeline Expectations

The "2025 is the year of agents" narrative is dangerously optimistic. Karpathy's experience with self-driving—a perfect demo in 2013, still being perfected in 2025—offers a sobering parallel. Plan for a decade-long transformation, not a two-year disruption.

2. Augmentation Over Automation

The highest-value opportunities lie in "Iron Man suits" (augmentation tools) rather than "Iron Man robots" (full automation). Products that enhance human capabilities while maintaining human oversight will dominate the next five years.

3. Infrastructure Plays

Just as the internet boom created value in infrastructure companies, the AI transformation will reward those building the scaffolding for agent-based computing.

4. Cross-Paradigm Competence

Winners will fluidly navigate between Software 1.0, 2.0, and 3.0, choosing the right tool for each component. Pure-play AI companies may struggle against those with hybrid approaches.

The Path Forward: Strategic Imperatives

For product builders and investors navigating this transformation, several principles emerge:

Build for the Verification Loop Every product decision should optimize the speed and accuracy of human verification. This isn't a temporary constraint—it's a fundamental design principle for the next decade.

Embrace Partial Autonomy Products should ship with an "autonomy slider," allowing users to dial up AI involvement as the technology matures and trust develops.

Invest in Infrastructure Adaptation Whether building or investing, prioritize companies that understand agents as first-class digital citizens requiring purpose-built infrastructure.

Maintain Historical Perspective We're witnessing the birth of a new computing paradigm. Like the transition from mainframes to PCs, the winners won't be determined by who moves first, but by who best understands the fundamental shifts underway.

Conclusion: The 1960s of a New Era

Karpathy's framework reveals that we stand at an extraordinary inflection point. For the first time in 70 years, the fundamental nature of software is changing. We have three distinct programming paradigms, a new class of digital infrastructure consumers, and the democratization of software development through natural language.

The contrast between Jace and Cora exemplifies why understanding these principles matters. Jace, with its Cursor-inspired interface and tight verification loops, has users reporting hours saved weekly. Cora, despite its elegant concept of email-as-narrative, struggles against the fundamental constraint of broken verification loops. The market's verdict is clear: AI products succeed when they enhance human agency, not when they abstract it away.

Yet we must temper our enthusiasm with realism. We are in the 1960s of this new computing era—a time of immense possibility but also fundamental limitations. The organizations that thrive will be those that understand both the transformative potential and the current constraints of these technologies.

For investors, the implications are stark: evaluate AI companies not by their ambitions but by their understanding of human-AI collaboration. Look for products with clear autonomy sliders, purpose-built interfaces for verification, and infrastructure designed for agent interaction. Avoid those chasing full automation without establishing the foundational loops of augmentation.

For builders, the message is equally clear: your north star should be the speed and accuracy of the human verification loop. Every design decision should ask: does this make it easier for humans to verify and guide AI output? Does our infrastructure speak to both human and agent consumers? Are we building Iron Man suits or chasing Iron Man robots?

The future belongs not to those who chase fully autonomous agents, but to those who thoughtfully build the bridges between human intelligence and artificial capability. In this new landscape, understanding Karpathy's framework isn't just helpful—it's essential for anyone serious about building or investing in the AI-powered future.

The question isn't whether this transformation will reshape every industry—it's whether you'll be among those who understand it deeply enough to capture the value it creates.

The Three Great Computing Transformations

My career has spanned three distinct computing eras, each lasting roughly two decades and characterized by iconic breakthrough applications that made abstract technology tangible to millions:

1980-2000: The PC/Internet Era
The transformation began with standalone computing, exemplified by Lotus 1-2-3, which made spreadsheets indispensable to financial professionals. Netscape later democratized internet access, connecting these isolated productivity islands. For those in finance, imagine working without spreadsheets—it's nearly impossible. That's the scale of transformation we're discussing.

2000-2020: The SaaS/Mobile Era
AWS eliminated infrastructure barriers with credit-card simplicity, while Apple's iPhone put computational power in every pocket. Real-time data became ubiquitous, fundamentally changing how investment decisions were made and communicated.

2020-2040: The Generative AI Era
NVIDIA's GPUs made large-scale AI computation feasible, while OpenAI's ChatGPT achieved the fastest technology adoption in history. But here's what's different: the pace of evolution within this wave dwarfs anything we've seen before.

The Unprecedented Velocity of AI Development

The numbers tell a story of exponential acceleration that defies historical precedent. In 2017, we had perhaps two significant AI models. By 2024, over 100 major models compete for attention, with thousands more on platforms like Hugging Face—a compound annual growth rate of 167%.

But model proliferation barely scratches the surface. At Google I/O, the company revealed that token usage had exploded 50-fold to 480 trillion tokens monthly. While user growth was impressive at 10x, usage grew 50x—indicating a fundamental shift from shallow interactions to deep, transformative engagement.

This explosion isn't just about scale; it's about capability evolution happening along three independent axes:

1. Pre-training scaling: Broader foundational knowledge through massive data ingestion

2. Post-training refinement: Specialized expertise for specific domains

3. Test-time computation: The game-changing ability to reason through problems

The February 2025 launch of reasoning models like DeepSeek marked an inflection point. Suddenly, AI wasn't just pattern-matching—it was thinking step-by-step through complex problems, showing its work, and distinguishing correlation from causation.

Software Development: The Canary in the Coal Mine

To understand investment management's future, observe software development's present. Cursor, an AI-enhanced IDE, rocketed to $100 million in annual recurring revenue faster than any SaaS company in history. The reason? Developers aren't just getting code suggestions—they're delegating entire tasks to AI.

OpenAI's Codex and Google's Jules represent the vanguard of this transformation. These aren't assistive tools; they're autonomous agents that execute complete coding tasks. Microsoft now generates 30% of its code through AI, heading toward 60% within two years. When coding—one of humanity's most complex cognitive tasks—can be delegated to AI, what makes investment analysis immune?

The numbers are staggering: 7 million developers building with Google's Gemini (5x growth in one year), enterprise AI usage up 40x, and GitHub repositories dedicated to AI growing vertically after ChatGPT's launch. This isn't experimentation; it's production-scale transformation.

The Abundance of Intelligence

What we're witnessing is an unprecedented abundance of intelligence capabilities. Every day brings new models, tools, and techniques. The old paradigm of building and protecting proprietary systems cannot compete with this explosion of innovation.

Consider the cost dynamics: AI processing costs have dropped faster than any technology in history—80% reduction in 24 months while capabilities expanded by orders of magnitude. A top-tier AI model now costs less monthly than a single Bloomberg terminal.

This creates a fundamental strategic choice for investment firms:

The Enhancement Path: Legacy providers like Bloomberg and Refinitiv bolt AI features onto existing systems, achieving 20-30% productivity gains while preserving current business models.

The Transformation Path: AI-native platforms rebuild workflows from scratch, achieving 10x capability multiplication through continuous learning systems and architectural advantages.

From Information Access to Autonomous Workflows

The evolution of AI in investment management follows three distinct horizons:

Horizon 1 - Information Access (Current State)
Most firms operate here, using AI for document summarization, semantic search, and basic automation. Productivity gains range from 30-75%, primarily in time savings on routine tasks.

Horizon 2 - Insight Generation (Emerging)
Leading firms are beginning to use AI for pattern detection across multiple data sources, integrating structured and unstructured data to generate novel insights. Each senior professional effectively gains the output of 1-2 junior analysts.

Horizon 3 - Autonomous Workflows (12-24 Months)
The future state features AI agents that autonomously monitor markets, execute analysis, and manage investment workflows 24/7. This isn't science fiction—it's the logical extension of capabilities already demonstrated in software development.

A Personal Case Study in Transformation

During Christmas last year, I faced the perennial challenge of writing a comprehensive investment memorandum—without associates to delegate to. Using Claude 3.5 and a reimagined workflow, I transformed a multi-week process into a matter of hours.

The key wasn't just using AI as a writing assistant. I fed it questionnaires, virtual data room documents, presentation materials, and crucially, transcripts from every due diligence meeting. These transcripts represented institutional knowledge that traditionally lived only in human memory. The AI synthesized everything into a 60-page institutional-quality memorandum that required minimal editing.

This wasn't a 30% productivity gain—it was a 10x transformation in how core work gets done.

The New Competitive Imperative

Major sovereign wealth funds now require sophisticated AI integration as a precondition for new allocations. This isn't optional enhancement; it's existential requirement. The investment industry faces the same choice that confronted software companies when the internet arrived: transform or become irrelevant.

The building blocks for transformation already exist:

• Reasoning models that work through complex investment theses

• Retrieval systems that access proprietary data

• Agent frameworks that execute autonomous analysis

• Code generation that overcomes LLMs' quantitative limitations

Success no longer comes from building proprietary AI systems—it comes from orchestrating the abundance of available intelligence better than competitors.

Strategic Imperatives for Investment Leaders

1. Embrace Abundance Thinking
The old model of protecting proprietary analytical methods cannot compete with daily innovation from millions of developers. Your competitive advantage lies in orchestrating external capabilities with deep domain expertise.

2. Start Tonight, Not Tomorrow
Every month of delay widens the gap with early adopters. Begin with simple experiments using Claude or GPT-4 for real analysis. The cost is trivial; the learning is invaluable.

3. Prepare for Role Transformation
Portfolio managers won't be replaced by AI—they'll orchestrate AI agent swarms that analyze thousands of opportunities simultaneously. The question isn't whether this will happen, but whether you'll lead or follow.

The Choice Before Us

In two years, investment professionals' jobs will be radically different. Not five years—two. The firms that capture the emerging abundance of intelligence will operate at 10x the analytical capacity of those protecting traditional processes.

When I exchanged emails with conference organizers to arrange my presentation, those emails were AI-generated—I merely approved them. This isn't the future; it's the mundane present. The extraordinary future lies in AI agents that will transform investment management as fundamentally as spreadsheets transformed finance forty years ago.

The only question that matters: Will you be among those doing the transforming, or among those being transformed?

Michael Bruck is Managing Partner at 71 Capital Ventures, where he leads AI strategy and implementation. This article is adapted from his presentation at the Robeco Investment Day, June 2024.

What makes this moment particularly significant isn't merely the integration of new technology, but a fundamental architectural transformation in how investment intelligence operates. This isn't simply about adding AI features to existing platforms; it's about reimagining the entire infrastructure of financial analysis and decision-making.

The Three Horizons of Investment Intelligence

Through my analysis of the rapidly evolving AI landscape, I've identified three distinct horizons that map the evolution of artificial intelligence in investment management:

Horizon 1: Information Access (Present Day)

Most investment firms today operate in this initial phase where AI primarily serves as an enhanced information processing system. Teams leverage tools that summarize earnings calls, extract data from filings, and organize market intelligence—valuable capabilities, yet fundamentally limited. The technology functions as a bolt-on feature to legacy platforms rather than a reimagined architecture.

The productivity gains here are non-trivial—analysts report 30-75% reductions in data retrieval time and approximately 4-5 hours of weekly time savings—but they represent evolutionary rather than revolutionary advancement. A Bloomberg Terminal with AI-enhanced search remains, at its core, the same analytical paradigm with faster information retrieval.

Horizon 2: Insight Generation (Emerging)

We now stand at the threshold of a more transformative phase where AI evolves from information organization to active insight generation. The foundational technologies enabling this shift include:

1. Internet-scale semantic search, which transforms the entire web into a queryable financial database

2. Model Context Protocol (MCP) implementations that seamlessly connect AI systems with disparate data sources

3. Multi-modal analysis frameworks that integrate text, numeric data, and visual information

The fundamental breakthrough in this horizon is that AI begins to identify patterns across diverse information sources and generate hypotheses autonomously. Early implementations suggest productivity equivalents of adding 1-2 junior analysts to each senior team member.

Yet this remains an intermediate stage in the evolution of investment intelligence.

Horizon 3: Autonomous Workflows (Future)

The ultimate destination—still several years away but clearly visible on the technological horizon—involves a paradigm shift from tools to autonomous systems. This third horizon will introduce:

1. Self-directing agent networks that continuously monitor markets without human activation

2. Causal reasoning engines that understand not just correlations but true cause-effect relationships

3. Adaptive knowledge frameworks that learn from institutional experience

The theoretical potential is staggering: a 5-10x increase in analytical capabilities with 24/7 market monitoring and compounding institutional knowledge advantages. This isn't about replacing human judgment but exponentially expanding its reach and depth.Software Development: The Canary in the AI Coal Mine

Code to Capital: How Software's AI Adoption Forecasts Investment Management's Future

The software development industry provides a compelling preview of the transformation now beginning in investment management. What makes this parallel so instructive is how rapidly software development has progressed from experimental AI implementations to mission-critical operational systems.

The velocity of this transition is perhaps best captured by the meteoric rise of AI-native development platforms. Cursor, for instance, scaled from $1 million to $100 million in annual recurring revenue in just two years—a growth trajectory that outpaces even OpenAI’s ChatGPT. This isn't merely a story of venture capital exuberance, but of fundamental productivity transformation that creates immediate, measurable value.

As Andrej Karpathy, former Tesla AI director, aptly described the shift in a viral X post: "There's a new kind of coding I call 'vibe coding,' where you fully give in to the vibes, embrace exponentials, and forget that the code even exists." This represents more than hyperbole—it captures how AI is fundamentally altering the interaction model between knowledge workers and their craft. The technology enables professionals to operate at a higher level of abstraction, focusing on intent and outcomes rather than implementation details.

Most critically for investment management, software development's AI transformation demonstrates that knowledge-intensive work—previously considered immune to significant productivity disruption—can experience order-of-magnitude efficiency gains when properly restructured around AI-native architectures. Just as developers now accomplish in hours what previously required days, investment analysts are poised for similar productivity leaps.

This transition follows a consistent pattern: initial adoption focused on basic assistance tools, followed by collaborative insight generation, and ultimately advancing toward autonomous capabilities. The compression of this adoption cycle in software development—from early experimentation to mainstream deployment in less than 18 months—suggests that investment management's transformation may unfold far more rapidly than conventional wisdom anticipates.

The investment firms that recognize this parallel and accelerate their own architectural transitions will likely establish sustainable competitive advantages, while those that view AI as merely an incremental enhancement risk falling irretrievably behind.

Understanding the Architectural Imperative

To appreciate why this transformation runs deeper than feature enhancement, we must recognize the historical parallels to previous computing transitions.

Just as the PC era demanded fundamentally different architectures than mainframe computing, and cloud computing required rethinking the entire infrastructure stack beyond traditional server farms, AI-native architectures demand a ground-up reimagining of how investment intelligence operates.

Consider the limitations of simply adding AI features to legacy platforms: enhanced search in a traditional terminal might deliver 20-30% productivity improvements at best. By contrast, systems designed around AI capabilities from inception can deliver the 10x productivity leap that represents true transformation.

The Emerging Team Models

As AI capabilities advance, investment teams are evolving toward two distinct but complementary human-AI collaboration patterns that I've termed the "factory" and "artisan" models:

The "Factory" Model

In this approach, autonomous AI agents handle predictable, routine processes like monitoring financial metrics, screening securities, or extracting data from filings. Human supervision remains, but the primary workflow is systematic and scalable. Early implementations have shown teams completing analytical tasks in half the traditional time through AI orchestration.

The "Artisan" Model

Here, AI functions more as an assistant augmenting experienced investment professionals in scenarios requiring judgment and creativity—complex portfolio construction, novel investment thesis development, or emerging market analysis. The AI serves as a thought partner, helping generate ideas and test hypotheses without replacing human decision-making.

What's particularly interesting is how the most sophisticated investment firms are strategically blending these approaches based on the specific workflow. Systematic monitoring might leverage the factory model, while thesis development follows the artisan approach—with AI seamlessly bridging both modes.

Real-World Transformation: 71 Capital Case Studies

The abstract potential of AI in investment management becomes concrete when examining implementations we've pioneered at 71 Capital:

Investment Memorandum Creation

At 71 Capital, one of our most successful AI implementations has been our structured document creation system for investment memorandums. The traditional process for generating these large and detailed documents was manual and labor-intensive, requiring weeks of analyst time to compile information from various sources—questionnaires, virtual data room documents, management presentations, meeting transcripts, and deep research reports.

By implementing a workflow-based system utilizing Claude 3.5 Sonnet alongside a programmatic workflow editor, we've achieved an order-of-magnitude reduction in production time. The system seamlessly synthesizes diverse inputs into structured, coherent, and high-quality investment documents for our Investment Committee.

What's particularly noteworthy is that this represents a workflow rather than an agent approach. The process demands predictable outcomes and standardized formats—exactly the type of structured, consistent process where a workflow approach shines. An agent, with its emphasis on autonomy, would be less suitable for this task where format adherence and predictability are paramount.

Customized Intelligence Newsletters

Another transformative implementation at 71 Capital has been our customized intelligence newsletter system. This three-stage process perfectly illustrates the evolution from basic data gathering to sophisticated AI-powered synthesis:

First, we scrape and gather data from various websites and online sources, creating a diverse pool of raw and unstructured information. Next, scheduled automated cron jobs systematically extract, structure, and store this web data in an organized database, ensuring reliable and consistent updates.

The true breakthrough comes in the final stage, where we use Large Language Models to aggregate, synthesize, and summarize this structured data into concise newsletters. This has dramatically reduced daily reading time from approximately two hours to around 10 minutes—an efficiency gain that multiplied across our entire team represents thousands of hours annually redirected toward higher-value analysis.

Again, this implementation leverages a workflow rather than an agent approach because the data sources are known and predictable, ensuring consistent outputs and format adherence. The structured automation provides the standardization and reliability our investment process demands.

Research Workbench Platform

Perhaps our most sophisticated AI implementation at 71 Capital has been our Research Workbench platform, which represents an evolution beyond traditional workflows toward what we call a "pseudo-agent" architecture. Unlike our customized newsletter system, which follows a fixed, structured process designed for standardized outputs, the Research Workbench employs a fundamentally different approach.

This pseudo-agent dynamically selects from a diverse toolkit—including a RAG Knowledge Base, internet search capabilities, and specialized analytical instruments—based on the specific query at hand. The contextual intelligence of this system allows it to navigate complex, unpredictable information requests with remarkable flexibility.

What distinguishes this approach from a pure workflow is its adaptive nature. Rather than following predetermined pathways, the system evaluates each query independently, selects optimal information retrieval strategies, and synthesizes responses tailored precisely to the analyst's needs. This contextual awareness enables the platform to effectively handle the varied and complex questions that emerge during the investment research process.

The core architecture integrates several advanced components: a comprehensive RAG Knowledge Base containing our proprietary research and analysis, dynamic internet search capabilities that extend beyond our internal data, contextual tool selection mechanisms that identify the most appropriate analytical approach for each query, integration with OpenAI's ChatGPT (o3-mini), and multi-user collaboration features that allow our entire investment team to leverage and contribute to the platform's capabilities.

The impact has been transformative, significantly reducing hours of research into concise, insightful summaries that address precisely what our analysts need to know. This evolution toward more agent-like behaviors while maintaining the reliability of structured workflows represents exactly the kind of architectural innovation that defines the transition between our second and third horizons of AI implementation.

Deep Research Integration

Our most advanced AI implementation at 71 Capital is our Deep Research platform, which represents a further evolution of our Research Workbench into truly agent-based territory. This system addresses broader, more open-ended questions compared to structured tasks like Investment Memorandums, requiring less rigid frameworks and more adaptive methodologies.

The fundamental architecture follows an iterative research loop that begins with clearly defined research objectives and parameters for exploration depth and breadth. From there, the system enters a sophisticated research cycle:

1. Intelligent Query Generation: The agent autonomously formulates targeted queries aligned with research goals, adapting its approach based on emerging information.

2. Parallel Web Searches: The system concurrently searches across multiple sources to efficiently gather comprehensive information, operating across diverse data landscapes simultaneously.

3. Results Processing: More than mere collection, the system analyzes and synthesizes findings to identify actionable insights, connecting disparate information points into coherent narratives.

4. Recursive Exploration: Perhaps most impressively, the system continuously deepens research by iteratively refining and expanding queries based on insights, navigating research paths that human analysts might miss.

Throughout this process, the agent maintains contextual awareness, continuously incorporating new findings to ensure all subsequent actions remain aligned with the original research objectives. The output consists of structured, comprehensive markdown reports complete with actionable insights, references, and citations.

This approach dramatically differs from traditional research workflows in its flexibility and context-awareness. For example, when we tasked the system with researching statistical arbitrage strategies across centralized and decentralized exchanges, it autonomously generated a sophisticated 30-page analysis in just 10 minutes—work that would have required days with traditional research approaches.

The critical innovation in our Deep Research platform is its ability to handle complex, evolving contexts. The system excels at generating comprehensive analysis for technical subjects like algorithmically-driven trading strategies, beginning with well-formulated queries, proceeding through clarification dialogues, and eventually producing structured reports that include sophisticated elements like pseudocode for implementation.

As the agent ecosystem continues evolving, we've carefully tracked the emerging competitive landscape across leading LLM platforms. When Gemini 1.5 Pro first piloted deep research capabilities, it set high benchmarks by integrating tightly with Google's ecosystem. However, ChatGPT quickly established dominance in complex analytical tasks that appeal to technical professionals. More recently, Perplexity has carved out a niche through rapid, interactive insights, while You.com targets enterprise-grade research demands, and xAI's Grok has made headlines with its uniquely real-time, social data-driven approach.

These platforms differ significantly in performance characteristics—from Perplexity's rapid 2-4 minute processing to ChatGPT's more thorough 30-minute deep dives, from source coverage ranging from ChatGPT's limited 25 sources to You.com's impressive 400+ simultaneous source processing. Each platform has carved distinct specializations, whether in technical research, coding tasks, market analysis, or real-time trend monitoring.

This competitive innovation cycle is precisely why we've designed our systems to be platform-agnostic, allowing us to leverage each model's strengths while mitigating their individual limitations. The result is an adaptive research capability that significantly expands our analytical horizon while maintaining the critical human judgment that defines our investment philosophy.

The Vendor Landscape: A Strategic Decision Matrix

Investment firms now navigate a complex decision landscape regarding AI implementation, choosing between AI-native startups and enhanced incumbent platforms:

AI-First Platforms

A wave of specialized vendors has emerged, including:

• Perplexity for Finance: Combining conversational interface capabilities with authenticated financial data sources

• AlphaWatch: An "AI copilot for hedge funds" integrating external financial data with proprietary information

• Brightwave: An AI research assistant generating financial analysis by synthesizing earnings reports and filings

These pure-play AI platforms often introduce innovative features ahead of incumbents and cater to specific niches within the investment workflow.

Enhanced Incumbent Platforms

Traditional market data providers are actively embedding AI capabilities:

• Bloomberg: Developed BloombergGPT (a 50-billion parameter finance-trained LLM) and integrated AI-written earnings summaries

• FactSet: Launching generative AI search and conversational APIs

• S&P Global: Implementing AI-enhanced search and data integration across its platforms

Incumbents benefit from trusted data and compliance frameworks, while often struggling to fundamentally reimagine their architectures.

The emerging industry pattern isn't one vendor category displacing the other, but rather a complex ecosystem where both approaches address different aspects of the investment process. Many sophisticated firms maintain relationships with traditional data providers while selectively integrating specialized AI tools for specific use cases.

Strategic Implementation Roadmap

For investment organizations navigating this transformation, I recommend a phased approach that balances immediate gains with long-term architectural advantages:

Near-Term (12-18 months)

• Evaluate platforms based on their architectural roadmap toward AI-native capabilities

• Prioritize systems that support universal connectivity standards

• Build internal competencies in prompt engineering and AI workflow design

Medium-Term (18-36 months)

• Implement internet-scale semantic search capabilities

• Develop hybrid workflows combining human judgment with emerging AI capabilities

• Restructure research processes to leverage multi-modal data integration

Long-Term (3-5 years)

• Prepare organizational structures for autonomous AI workflows

• Design governance frameworks for agent-based systems

• Develop performance metrics for measuring AI-augmented analyst effectiveness

The Fundamental Thesis: AI as Force Multiplier

The central argument that I've been advancing to investment leaders is that AI represents not merely an incremental improvement but a fundamental force multiplier for investment professionals. When properly implemented through AI-native architectures, these technologies promise:

1. Comprehensive market intelligence spanning the entire web rather than just subscribed databases

2. Expansion of research coverage from dozens to hundreds of companies per analyst

3. Enhanced competitive intelligence through continuous monitoring of market trends

4. Accelerated investment thesis generation, reducing research cycles from weeks to hours

This force multiplier effect—potentially increasing productivity by an order of magnitude—represents a competitive imperative for investment firms. The organizations that recognize this architectural evolution and position themselves to capture the transformative benefits of truly AI-native systems will likely establish compounding advantages, while those implementing superficial AI features risk falling behind.

The journey to fully AI-native investment operations is still unfolding, but the trajectory is clear: AI is evolving from experimental technology to the fundamental infrastructure of modern investment management. Just as previous technological revolutions created clear winners and laggards, this AI transformation in investment management will likely reward the architecturally astute and strategically bold.

The semiconductor industry just witnessed a momentous event: Dylan Patel of SemiAnalysis and Doug O’Laughlin of Fabricated Knowledge announcing the merger of their research organizations during a podcast hosted by Ben Thompson on Stratechery. These two titans, each known for unparalleled insight into semiconductors, AI infrastructure, and the semiconductor supply chain, are now collaborating to drive SemiAnalysis to unprecedented heights. This merger is poised to reshape how we understand and analyze one of the world’s most critical industries.

Patel and O’Laughlin have cultivated strong individual followings. Patel, the outspoken analyst, has carved a niche through his comprehensive AI supply chain research, while O’Laughlin’s focus has been on the financial and technological implications of the semiconductor space. Together, their collaboration represents a perfect marriage of technical and financial perspectives—one that promises even more incisive analysis of semiconductors’ future.

Why This Matters

The merger isn’t just about joining two Substack powerhouses. It reflects the larger transformations underway in the semiconductor and AI industries. SemiAnalysis is much more than a newsletter: it has grown into a comprehensive research and consulting powerhouse. Patel shared that the company’s offerings now include tracking global data centers, fabs, accelerator shipments, and semiconductor equipment movements. With O’Laughlin on board, this expansion is accelerating at a pace Patel likens to a “rocket ship.”

The combined reach of SemiAnalysis now spans major tech companies, hedge funds, private equity firms, and AI labs. SemiAnalysis is no longer simply reporting on trends—it is shaping them. As O’Laughlin aptly noted, if you can’t beat them, join them. His decision to merge reflects a recognition that their joint vision is stronger than either could achieve alone.

Key Insights into the Industry’s Future

Peak Lithography and Bottlenecks

One of the central insights from the podcast revolved around “peak lithography.” As extreme ultraviolet (EUV) technology stabilizes, future innovation is shifting away from traditional lithography techniques toward advancements like gate-all-around transistors and 3D architectures. This signals a broader shift in semiconductor priorities, with memory technologies, High Bandwidth Memory (HBM), and packaging gaining prominence.

O’Laughlin’s perspective on these developments underscores that the era of pure lithographic scaling is giving way to innovations in process control, advanced etching, and packaging. Patel added that these shifts mark a second wave of semiconductor advancements, fueled by AI’s immense compute demands.

AI’s Transformation of the Semiconductor Landscape

AI is central to Patel and O’Laughlin’s thesis on industry disruption. They predict the AI supply chain will generate hundreds of billions of dollars annually, driving innovation at every layer—from materials to advanced packaging. Patel argued that the horizontal structure of supply chains has created inefficiencies, which SemiAnalysis is uniquely equipped to address.

Both analysts see AI as not just a disruptor but the next defining wave of semiconductors. They noted that memory bottlenecks and cost scaling in AI workloads are now as critical as transistor density. Patel highlighted the massive infrastructure buildouts underway for AI and their transformative effect on the industry. He anticipates growing investment in memory, packaging, and 3D stacking, ensuring AI’s dominance in semiconductor strategy for years to come.

Geopolitics and the Rise of China

The analysts also touched on geopolitics, particularly China’s ambitions in the semiconductor space. Patel described China’s “decentralized” approach to developing its semiconductor capabilities by simultaneously deploying domestic and foreign tools. This iterative strategy aims to catch up with Western technology in record time. O’Laughlin cautioned that China’s focus on becoming self-reliant poses both a challenge and an opportunity for the West. As the U.S. prioritizes AI and semiconductor innovation, maintaining its lead will depend on robust industrial policy and continuous technological breakthroughs.

Intel and the Struggle for Relevance

Intel was a focal point in the discussion. The analysts acknowledged Intel’s storied history of innovation but raised concerns about its ability to recover from years of missteps. Patel noted that Intel is attempting to regain leadership with technologies like gate-all-around transistors and backside power delivery. However, its cultural inertia and lack of strategic focus make recovery an uphill battle.

O’Laughlin floated the possibility of a bold restructuring—splitting Intel’s fab and design divisions to prioritize its foundry ambitions. He suggested that embracing a pure-play foundry model could allow Intel to compete more effectively with TSMC. Yet, as both analysts noted, this transformation requires leadership willing to “burn the ships” and commit fully to a new vision.

Closing Thoughts

The merger of SemiAnalysis and Fabricated Knowledge represents a watershed moment in semiconductor analysis. By combining their talents, Patel and O’Laughlin are poised to deliver groundbreaking insights into an industry at the heart of global innovation and geopolitics. From the AI-driven transformation of the semiconductor supply chain to the strategic challenges facing legacy giants like Intel, their unified voice will shape how investors, policymakers, and technologists navigate this dynamic landscape.

The semiconductor and AI industries are converging in ways that will define the future of technology and global innovation. For readers eager to stay ahead of these groundbreaking developments, the merger of SemiAnalysis and Fabricated Knowledge is a pivotal moment. To dive deeper into the insights shaping these industries, I highly recommend following and subscribing to Stratechery, Fabricated Knowledge and SemiAnalysis. Their unparalleled analysis will keep you informed about the critical trends and strategies driving the next wave of technological transformation.

Recent developments have cast Intel’s future in a new light. Reports have surfaced that Qualcomm approached Intel with a potential takeover offer, signaling a pivotal moment for both companies and the broader U.S. semiconductor industry. This potential deal, which could be one of the largest in tech history, comes as Intel faces the most severe crisis in its five-decade history, and Qualcomm looks to expand its dominance beyond its traditional markets.

Intel’s Struggles and Qualcomm’s Ambitions

Intel has been grappling with significant challenges, including strategic missteps, a delayed response to the AI boom, and a costly turnaround effort under CEO Pat Gelsinger. The company, once the world leader in both chip design and manufacturing, has lost substantial ground to rivals like TSMC and Samsung, particularly in the advanced node space critical for AI and other high-performance applications.

Qualcomm, on the other hand, is a key player in mobile chips and has recently expanded into automotive and IoT markets. While it outsources its manufacturing to companies like TSMC, acquiring Intel would allow Qualcomm to diversify into the PC and server markets, as well as potentially gain a foothold in semiconductor manufacturing. However, it’s uncertain whether Qualcomm would retain Intel’s fabs, as managing such capital-intensive operations could be a heavy burden for a company that has traditionally focused on design rather than manufacturing.

The Geopolitical and Economic Context

Against the backdrop of a potential Chinese invasion of Taiwan, the U.S. semiconductor industry is at a critical juncture. U.S. companies like Qualcomm, Nvidia, and Apple rely heavily on TSMC for the production of their most advanced chips. Should Taiwan’s semiconductor manufacturing capabilities be disrupted, these companies—and the broader U.S. tech industry—would face a severe supply chain crisis.

Intel’s fabs, while lagging behind TSMC and Samsung in advanced node production, represent the only domestic U.S. option for semiconductor manufacturing at scale. With government subsidies through the CHIPS Act, Intel has been attempting to revitalize its fabs and position itself as a viable alternative to overseas manufacturing, but its progress has been slow. Qualcomm’s potential acquisition of Intel, however, raises the question of whether U.S. tech giants could pool resources to bolster Intel’s fabs and secure a domestic manufacturing base.

A New Strategic Direction for Intel?

If Qualcomm were to acquire Intel, a likely scenario would involve Qualcomm divesting parts of Intel’s business. This could include selling off Intel’s chip design division to eliminate competitive concerns, while keeping the foundry business intact. A consortium of U.S. tech giants, including Apple, Nvidia, and Amazon, could then invest in Intel’s manufacturing operations, ensuring they have a secure, U.S.-based supply chain for advanced chips.

Such a move would be transformative for the U.S. semiconductor industry. It would align the interests of U.S. tech companies, which are currently reliant on TSMC, with the national security imperative of securing a domestic semiconductor manufacturing capability. Qualcomm, by divesting Intel’s design business and focusing on its traditional strengths, would ease competitive tensions while enabling Intel’s fabs to be the focus of an industry-wide effort to compete with TSMC and Samsung.

The Importance of Fabs in U.S. Tech Sovereignty

The potential Qualcomm-Intel deal underscores the growing importance of semiconductor manufacturing to U.S. economic and national security interests. If China were to cut off access to Taiwanese fabs, the U.S. would need a domestic option to produce the advanced chips required for everything from smartphones to data centers and AI applications. Intel’s fabs are currently the only viable option to fill that gap, but they need significant investment to become competitive with the likes of TSMC.

With the right backing—both from the government and from U.S. tech giants—Intel’s fabs could become the cornerstone of a domestic semiconductor manufacturing ecosystem. Qualcomm’s approach to Intel may have opened the door to a broader industry-wide realignment that could ensure U.S. tech sovereignty in the face of increasing geopolitical risks.

The Path Forward

The Qualcomm-Intel discussions mark a turning point for both companies and for the U.S. tech industry. Intel’s future is uncertain, but this deal could offer a way forward. By divesting its design business and focusing on its manufacturing operations, Intel could position itself as the U.S. answer to TSMC. With the backing of Qualcomm and other tech giants, Intel’s fabs could finally become the competitive force that the U.S. semiconductor industry so desperately needs.

The semiconductor industry is facing unprecedented challenges, and Intel’s fate could have far-reaching implications. Qualcomm’s potential acquisition could be the catalyst that reshapes the industry, securing U.S. leadership in semiconductor manufacturing for the next decade and beyond.

In the fast-paced world of technology, the spotlight is always shifting. The meteoric rise of generative AI tools like ChatGPT 4o, Sora, CoPilot+ and more, has captured the public imagination and dominated industry discourse. The metaverse, once the darling of the tech press and the obsession of Silicon Valley, has been totally overshadowed in this AI spring. Headlines proclaiming the metaverse as the next big thing have been superceded by eager coverage of the transformative potential of large language models and AI image generators.

But while the spotlight may have moved on, the work of building the metaverse continues apace. Behind the scenes, away from the fickle glare of the media, companies large and small are still pouring billions into the foundational technologies and platforms that will enable the immersive internet. The metaverse, it seems, is not dead - it's just in a chrysalis stage, quietly metamorphosing into its next incarnation. And when it emerges, it will likely leverage the power of the very AI systems that stole its thunder.

The concept of the metaverse, a term coined by science fiction author Neal Stephenson in his 1992 novel Snow Crash, has captured the imagination of Silicon Valley in recent years. The idea of a shared, immersive, and persistent virtual space that mirrors and augments our physical reality has been the stuff of science fiction for decades. But now, thanks to rapid advancements in gaming, virtual reality, blockchain technology, artificial intelligence, and 3D graphics, the metaverse is increasingly being seen as the next major computing paradigm - and potentially the future of the internet itself.

Two influential thinkers are at the forefront of this vision: Chris Dixon, a general partner at venture firm Andreessen Horowitz and a leading investor in web3 and gaming startups, and Matthew Ball, a strategist, writer and the former global head of strategy for Amazon Studios. Both have been at the forefront of theorizing and evangelizing the potential of the metaverse and the key technologies that will power it.

The Metaverse Masterminds: A Shared Vision for the Internet's Immersive Evolution

I recently listened to two separate interviews of Dixon and Ball where they laid out remarkably aligned visions for what they call the next generation of the internet. This future web will be immersive, spatially-native, and imbued with the decentralized ethos of web3. It will be a network of interconnected virtual worlds where our digital identities and assets move freely, unconstrained by the walled gardens of today's dominant platforms. And artificial intelligence, they believe, will also play a critical role in bringing this vision to life.

"The metaverse to me is essentially the internet in 3D," says Ball. "Not just virtual worlds experienced in a headset, but an evolution of the internet from a primarily text and image-based medium to one centered around interactive 3D spaces and experiences." Dixon similarly describes a future internet that is richer and more engaging, "an embodied internet where you're in the experience, not just looking at it."

But perhaps more important than the immersive nature of the metaverse is its underlying architecture and incentive structures. And this is where web3 principles come into play. Dixon and Ball both see blockchain technology, with its ability to enable decentralized ownership, provenance, and scarcity of digital assets, as a fundamental building block of the open metaverse.

"Blockchains allow building virtual worlds with the advanced capabilities of centralized platforms but where users and creators have rights and agency," Dixon explains. "They can own their identity, their audience, their creations. That's a big shift from web2 and it will lead to an explosion of innovation." Ball echoes this sentiment, highlighting the potential for web3 to power an economy of interoperable virtual goods and services across the metaverse.

Gaming, in particular, is seen by both Dixon and Ball as a key onramp into this new internet. Multiplayer games like Fortnite, Minecraft, and Roblox have already introduced hundreds of millions of users to shared virtual worlds. And the rise of web3 gaming, with its player-owned economies powered by NFTs and cryptocurrencies, points to a future where our virtual lives are just as meaningful and valuable as our physical ones.

But it's not just about the economics. Ball sees AI as playing a crucial role in making these virtual worlds more dynamic and engaging. "Generative AI tools could potentially help lower content creation costs and allow for more rapid iteration and personalization of game worlds," he notes. However, he predicts that the biggest gaming franchises, with their vast data and resources, will benefit disproportionately from this technology.

Dixon takes this idea even further, envisioning AI characters that are indistinguishable from human players. "Large language models, paired with the ability to query vast knowledge graphs, will allow non-player characters to engage in freeform dialogue," he wrote in a blog post. These AI agents could serve as guides, collaborators, or even central characters in immersive stories and experiences.

More broadly, AI is seen as a key enabling technology for making the metaverse more accessible, adaptive, and engaging. Intelligent systems could optimize virtual world rendering, power realistic avatars, and surface hyper-relevant content based on our interests and context. But this also raises important ethical questions around the persuasive power of these systems and the need to keep them accountable and aligned with human values.

Despite the clear potential, both Dixon and Ball acknowledge the very real challenges that stand in the way of this grand vision becoming reality. The technology is still maturing. Consumer adoption of VR headsets remains low. And the user experience of web3 products often leaves much to be desired. There are also thorny questions around interoperability, regulation, and sustainable business models for the open metaverse.

But both remain unwavering in their conviction that, as Ball puts it, "we are on the cusp of a new era in which the internet isn't something we reach for, or that runs underground and is transmitted through the air. Instead, it will be all around us and we will be in it." The building blocks, from blockchain to AI to spatial computing, are falling into place. The exact shape of our metaverse future may still be hazy, but its potential to transform nearly every aspect of human life is coming into focus.

Digital Dreamscapes: Why the Metaverse Feels Like a Flashback to Web 1.0

As I listened to Dixon and Ball lay out their expansive visions - at times sounding more like science fiction writers than tech executives - I could not help but be reminded of the early days of the internet. In the 1990s, evangelical technologists spoke of a global, democratizing information network that would transform nearly every aspect of society. At the time, their claims sounded outlandish. Looking back, if anything, they were too conservative.

Dixon and Ball may well be the metaverse's equivalent of those early internet pioneers - the architects sketching out a new virtual world that, to paraphrase William Gibson, is already here, just not evenly distributed. Whether the metaverse ends up evolving as they predict remains to be seen. But it's undeniable that something big is brewing at the intersection of gaming, web3, AI and spatial computing. And if the past is any indication, we may look back at this moment as the start of a technological transformation even more profound than the dawn of the web. The future internet is coming. And it will be intelligently immersive.

The world is witnessing a profound shift in the global economic order, driven by the forces of geoeconomic fragmentation (GEF). As trade tensions escalate and geopolitical rivalries intensify, the once seemingly inexorable march of globalization has given way to a new era of economic realignment. The United States and China, the world's two largest economies, find themselves at the epicenter of this tectonic shift, as they seek to decouple their deeply intertwined supply chains and assert their technological primacy. Yet, as the old order crumbles, new opportunities arise for countries that can navigate the fault lines of this fractured landscape and emerge as the connective tissue of the 21st-century global economy.

The evidence of GEF is mounting, even if the headline numbers on global trade and investment flows have not yet reflected a sharp decline. Underneath the surface, a "great reallocation" is underway, as companies rethink their supply chains and governments erect new barriers to trade and investment (Alfaro and Chor, 2023). The share of bilateral trade between geopolitically distant countries has started to decline relative to trade within allied blocs, a trend that has accelerated since the onset of the Russia-Ukraine war (Gopinath et al., 2024). The countries caught in the middle of this decoupling, however, face a new set of opportunities and risks.

During the Cold War era, the global economy was neatly divided into two rival blocs, with little room for maneuver for the nonaligned countries. Trade between the Western and Eastern blocs declined precipitously, while trade with nonaligned economies remained limited. In contrast, the current wave of GEF is unfolding in a much more integrated global economy, with complex supply chains and interdependencies that defy simple bifurcation. As a result, we are seeing the emergence of a new category of "connector" countries that are stepping in to bridge the gaps between the rival blocs (Freund et al., 2023).

These connector countries, such as Vietnam and Mexico, are benefiting from the reallocation of trade and investment flows away from China and toward more politically aligned partners. As Gopinath et al. (2024) show, the countries that have gained the most in U.S. import market share are also the ones that have seen the biggest increases in Chinese exports and outward foreign direct investment. This suggests that even as the direct links between the U.S. and China are severed, indirect ties via third countries are being strengthened.

However, the ability of connector countries to play this role is not unlimited, particularly when it comes to the flow of talent and advanced technologies. As concerns over national security and technological rivalry intensify, the barriers to the diffusion of cutting-edge technologies across borders are likely to grow. Export controls, investment screening mechanisms, and visa restrictions are already being deployed to prevent sensitive technologies from falling into the hands of geopolitical rivals. In this context, the countries that can position themselves as trusted intermediaries for the flow of talent and technology will have a unique advantage.

Enter the United Arab Emirates (UAE), a small but strategically located country that has long punched above its weight in the global economy. With its business-friendly policies, open economy, and neutral foreign policy stance, the UAE is well-positioned to emerge as a key connector country in the era of GEF. The UAE's leadership has made no secret of its ambitions to become a global hub for innovation and technology, investing heavily in emerging sectors such as artificial intelligence, blockchain, and space exploration.

What sets the UAE apart from other potential connector countries is its ability to attract and retain global talent. Through initiatives such as long-term golden visas and reforms to its citizenship laws, the UAE has made itself an attractive destination for highly skilled workers from around the world. This talent pool, combined with the UAE's strategic location and strong ties to major global economies, could make it a natural hub for the diffusion of advanced technologies.

Of course, the UAE is not without its limitations. Its relatively small domestic market and dependence on foreign talent could make it vulnerable to shifts in the global competition for skills. Its location in a volatile region also poses geopolitical risks that could deter some firms from locating sensitive R&D activities there. Nevertheless, the UAE's unique combination of strengths gives it a real shot at carving out a niche as a connector country for talent and technology flows.

As the world enters a new era of geoeconomic fragmentation, the countries that can adapt to the changing landscape and find ways to bridge the gaps between rival blocs will be the ones that thrive. Traditional trade and financial flows are already being realigned, but the next frontier will be the battle over talent and technology. In this context, the emergence of new connector countries like the UAE could be a defining feature of the 21st-century global economy. By providing a trusted platform for the exchange of ideas and innovations across borders, these countries could help to mitigate the worst effects of GEF and keep the flame of global cooperation alive. As policymakers and business leaders navigate this brave new world, they would do well to keep a close eye on the evolving role of connector countries in shaping the future of globalization.

References: Alfaro, L., & Chor, D. (2023). Global Supply Chains: The Looming "Great Reallocation". NBER Working Paper 31661.

Freund, C., Mattoo, A., Mulabdic, A., & Ruta, M. (2023). Is US Trade Policy Reshaping Global Supply Chains? World Bank Policy Research Working Paper 10593.

Gopinath, G., Gourinchas, P. O., Presbitero, A. F., & Topalova, P. (2024). Changing Global Linkages: A New Cold War? CEPR Discussion Paper DP18997.

In the rapidly evolving landscape of artificial intelligence (AI), deciphering when the current hype will translate into tangible revenue is a subject of much debate and anticipation. As we navigate through the crescendo of AI enthusiasm, marked notably by innovations like ChatGPT, a pivotal question arises: When and how will AI's promise convert into real economic value? A comprehensive analysis from Jefferies offers insightful perspectives into this transition, highlighting the pivotal role of generative AI products, the readiness of enterprises, and the strategic positioning of major tech players.

The Hype Cycle: Where Are We?

Generative AI, especially post the unveiling of ChatGPT, has sparked a new era of excitement and speculation in the tech world. This hype isn't without merit; AI's potential to redefine industries is palpable. However, as with any technological advancement, there's a cycle from the initial excitement to the realization of practical, revenue-generating applications. According to Jefferies, while the current buzz around AI is significant, the materialization of substantial revenues is likely to unfold from late 2024 into 2025.

This forecast is underpinned by a critical observation: the enterprise adoption of generative AI products is still in its nascent stages. Many enterprises are currently in the phase of exploring AI capabilities, figuring out use cases, and integrating AI into their operations in a way that justifies the cost. This exploratory phase is crucial, as it sets the groundwork for broader adoption and integration across industries.

The Path to Revenue: Key Players and Strategies

Jefferies' analysis points to several key players poised to benefit from the AI revolution. Microsoft (MSFT), with its strategic partnership with OpenAI, is highlighted as a frontrunner. This partnership, leveraging Azure OpenAI services, places Microsoft at the nexus of AI infrastructure and application development, making it a key beneficiary of the AI transformation. Amazon (AMZN) and Google (GOOGL), with their formidable cloud services (AWS and Google Cloud, respectively), are also well-positioned to capitalize on the growing demand for AI-powered solutions.

Furthermore, the report sheds light on the strategic importance of adopting a basket approach to investment in AI. This approach involves investing across a spectrum of companies involved in the 'picks and shovels' of AI—ranging from infrastructure providers to application vendors. This diversified strategy is recommended to capture the broad value AI is set to deliver across different sectors and functions.

The Road Ahead: Challenges and Opportunities

Transitioning from hype to revenue in the AI domain is not without its challenges. Issues such as data privacy, the cost of AI integration, and the need for enterprises to modernize their data infrastructure are among the hurdles to widespread adoption. However, these challenges also present opportunities for innovation and growth. Companies that can navigate these hurdles effectively, offering scalable, secure, and cost-efficient AI solutions, are likely to emerge as leaders in the AI era.

Moreover, the report anticipates a surge in demand for data governance solutions as companies prepare their infrastructure for AI deployment. This anticipates a broader trend where AI's integration into business operations will necessitate advancements in related domains such as cloud computing, data security, and modernization of IT infrastructure.

An Inevitable Revolution

The transition from AI hype to tangible revenue is a complex journey marked by technological, strategic, and operational milestones. While the timeline for this transition is subject to various factors, including enterprise readiness and technological advancements, the direction is clear: AI is poised to transform industries in profound ways. For businesses, staying ahead in this revolution will require not just investment in AI technologies, but a comprehensive strategy encompassing data management, security, and infrastructure modernization. As we move closer to realizing AI's full potential, the focus shifts from questioning if AI will transform industries to strategizing how to navigate and capitalize on this transformation effectively.

Photo credit: Kennevia Photography

I recently had the opportunity to speak at the Jefferies Asia Forum alongside Vikram Dewan, Chief Information Officer, and Conor O’Mara, Managing Director both from Jefferies, about how the firm is leveraging AI and natural language processing (NLP) to streamline workflows and unlock productivity for their analysts and portfolio managers. The key takeaway: AI is already delivering significant value and its impact will only grow in the years ahead.

Distilling Insights from Vast Amounts of Financial Documents

One of the most impactful applications is using AI to automatically generate summaries of the massive amounts of information investment professionals need to stay on top of, from earnings call transcripts to research reports. By distilling these documents into their key insights, AI can save analysts countless hours and ensure crucial details are not overlooked.

For example, using Anthropic’s Claude 3 Opus model, I was able to feed in both the transcript (created with Otter.ai) of a Jefferies conference session as well as the presentation deck, and the AI-generated a comprehensive summary, complete with key insights, actionable takeaways, and overall sentiment analysis - within seconds. This ability to synthesize information from multiple sources and formats is a capability that would take even the most skilled analyst substantial time and effort.

But beyond just summarization, these foundation large language models (LLMs) can be leveraged to build powerful research chatbots and knowledge bases. By ingesting and embedding a firm's research reports, meeting transcripts, and other proprietary data sources, these AI systems can instantly retrieve highly relevant information in response to an analyst's natural language queries - acting as an omniscient research assistant to augment the investment research process.

Enhancing LLMs with Contextual Knowledge for Investment Research using RAG

In my presentation at the Jefferies Asia Forum, I demonstrated a powerful system I created using an easy-to-use chatbot creation platform from GPTBots that leverages a technique called Retrieval Augmented Generation (RAG) to augment language models with highly relevant domain-specific information. The key idea behind RAG is to combine the broad general knowledge and powerful language generation capabilities of LLMs with a contextual retrieval system that can surface the most salient information from a substantial corpus of documents.

To showcase the potential of this approach for investment managers, I ingested a wide range of financial data sources into the GPTBots RAG system, including Jefferies research reports, earnings call transcripts, 10-K filings, news articles from Bloomberg and the Wall Street Journal, and even websites of private companies. By embedding this proprietary information into a vector database, the system can perform a semantic search to find the snippets of text that are most relevant to an analyst's query - going beyond simple keyword matching to understand the true intent behind the question.

The impact of augmenting the LLM with this curated domain knowledge is substantial. Rather than the model trying to generate an answer based solely on its training data - which may be outdated or biased towards more general information - it can now draw upon the expertise contained within the firm's own research to inform its outputs. This reduces the risk of hallucinations or inaccuracies and allows the model to provide highly targeted insights that are directly applicable to the companies and themes that the portfolio managers care about.

But beyond just serving as an information retrieval system, the integration of the LLM and the vector database through the RAG architecture enables powerful natural language interactions. Analysts can engage with the system much like they would converse with a highly knowledgeable colleague - asking follow-up questions, seeking clarification or additional context, and rapidly zeroing in on the key insights they need to inform their investment theses.

To run through a concrete example, let's say an analyst wanted to quickly get up to speed on the key takeaways from NVIDIA's latest GPU Technology Conference (GTC). Rather than having to manually read through multiple research notes or watch hours of presentation videos, they could simply ask the RAG system "What were the major announcements from NVIDIA's GTC conference this year?" The model would then search through its embedded knowledge base of conference transcripts, research reports, and news articles to surface the key details on NVIDIA's latest product and strategy announcements, generating a succinct summary that highlights the most important implications for investors as shown below.

During NVIDIA's GTC conference this year, several major announcements were made:

1. Blackwell GPU Architecture: NVIDIA unveiled the Blackwell GPU architecture, which offers significant improvements over its predecessor, particularly in terms of performance and power efficiency. The Blackwell GPUs are designed using a custom 4NP TSMC process, contain 208 billion transistors, and introduce six transformative technologies for AI training and real-time Large Language Model (LLM) inference for models scaling up to 10 trillion parameters.

2. Performance and Efficiency: The Blackwell architecture is said to deliver 20 PetaFLOPS of AI performance, which is 4x faster on AI Training workloads and 30x faster on AI Inferencing workloads compared to the H100 architecture. It also boasts up to 25x more power efficiency than its predecessor.

3. Second-Generation Transformer Engine: This technology within the Blackwell platform allows for 4-bit floating point AI calculations, doubling the compute performance and model sizes from the previous generation.

4. NVLink 5.0 Technology: The new NVLink technology enables up to 576 GPUs to be linked together at speeds up to 1.8 TB/sec, essential for complex AI models.

5. Grace Blackwell Superchip: This combines Blackwell GPUs with an Nvidia Grace CPU, offering a highly specialized system for demanding AI workloads.

6. Software Suite AI Enterprise 5.0 and Nvidia Inference Microservices (NIM): These ecosystem improvements simplify the deployment of AI models, making it easier for businesses to integrate AI capabilities into their operations.

7. Expansion of Applications and Industries: NVIDIA is working to bring critical GenAI compute solutions to various industries, including healthcare, heavy industry, automotive, robotics, manufacturing, telecommunications (6G), weather forecasting, and more.

8. Grace Hopper Superchip: NVIDIA showcased its Grace Hopper superchip, designed for AI applications, further cementing its commitment to the AI and GPU market.

These advancements are expected to solidify NVIDIA's position as a leader in the AI and GPU market, significantly widening its competitive moat.

Or if a portfolio manager wanted to dive deeper into a particular product line, they could ask a more targeted question like "What are the key differences between NVIDIA's H100 and A100 GPUs?" The system would again comb through its contextual knowledge to pull out the relevant performance and specification details, empowering the PM with the information they need to make faster and more informed investment decisions.

The key differences between NVIDIA's H100 and A100 GPUs are as follows:

1. Architecture: The H100 is based on the new NVIDIA Hopper architecture, while the A100 is based on the previous NVIDIA Ampere architecture.

2. Performance: The H100 delivers a significant performance leap over the A100. For instance, it provides 6x the peak compute throughput of the A100 for the most compute-hungry workloads.

3. Tensor Cores: The H100 features fourth-generation Tensor Cores that are up to 6x faster chip-to-chip compared to the A100's Tensor Cores. The H100 Tensor Cores also support the new FP8 data type, offering 4x the rate of A100's 16-bit floating-point options.

4. Memory: The H100 uses HBM3 memory, which is faster than the A100's HBM2, providing a substantial increase in memory bandwidth. The H100 offers up to 80 GB of memory size and a data rate of 2619 MHz DDR, compared to A100's 40 GB and 1215 MHz DDR.

5. Energy Efficiency: The H100 is designed to be up to 25x more power-efficient compared to its predecessor.

6. Compute Capability: The H100 supports Compute Capability 9.0, which introduces new features and improvements over the A100's Compute Capability 8.0.

7. Security Features: The H100 introduces new security enhancements and confidential computing features not present in the A100.

8. L2 Cache Size: The H100 has a larger L2 cache size of 50 MB compared to the A100's 40 MB.

9. Manufacturing Process: Both the H100 and A100 are manufactured using a 4N process customized for NVIDIA, but the H100 has 80 billion transistors compared to the A100's 54.2 billion, indicating a more complex and capable GPU.

10. TDP (Thermal Design Power): The H100 has a higher TDP of 700 Watts compared to the A100's 400 Watts, reflecting its higher performance capabilities.

These differences collectively contribute to the H100's superior performance in AI, HPC, and data analytics applications, making it a more advanced and efficient solution for complex computational tasks.

As I showed in my demo with the answers above, these are just a few examples of the power of augmenting LLMs with targeted information retrieval capabilities. By giving analysts the ability to quickly extract insights from the mountain of financial data that they are inundated with every day in natural language, RAG systems have the potential to dramatically boost the efficiency and effectiveness of the investment research process.

The Future of AI-Driven Investment Management: Opportunities, Challenges, and the Imperative to Adapt

To be clear, today's AI is not without limitations. These language models can hallucinate information, struggle with numerical analysis, and lack the domain expertise to independently drive investment decisions. But by strategically leveraging AI to augment human capabilities - automating the most laborious parts of the investment research process while still relying on the judgment of experienced PMs in the loop - the technology can be transformational in improving the productivity and quality of the investment management process.

Of course, successfully implementing AI in a regulated and hyper-competitive industry like investment management is not without challenges. But firms like Jefferies are showing that with the right approach, the benefits far outweigh the risks. As more and more investment managers embrace AI, those that fail to adapt will undoubtedly be left behind.

Looking to the future, we are just scratching the surface of AI's potential in the investment industry. As the technology behind language models continues to advance and we solve key challenges like incorporating real-time market data, the ability for AI to increase productivity to allow PMs to focus on creating alpha will grow exponentially. With the asset management industry's smartest minds increasingly being augmented with superhuman AI research assistants, in the future, technology-driven investment management will become the new normal, the same way that decades ago in the PC era, spreadsheets became an indispensable tool for portfolio managers.

Note: The article was written with the assistance of Claude 3 Opus based on an AI-generated transcript of my presentation.

The Revolution in AI

The origins of AI trace back to humble algorithms, which over time, evolved into the complex, generative, large language models we see today. These models, like OpenAI's GPT series, have capabilities that are nothing short of mind-blowing. From generating human-like text to creating art and music, the feats achieved today would seem like magic just a few years ago. Huang aptly coined it, "This is the iPhone moment for AI."

LLMs as a New Computing Platform

"Human is the new programming language," is a phrase that captures the essence of the new wave in computing. Gone are the days of cryptic code being the primary mode of interaction. Now, through natural language processing, a broader population can "speak" to computers. Large Language Models (LLMs) stand at the forefront of this transformation, with generative AI playing the role of the "killer app" for this paradigm.

Software 2.0

How AI Will Transform Development: Imagine an environment where developers no longer code but explain their ideas, and the AI realizes them. The concept of Software 2.0 stands on this foundation, where generative AI can synthesize code, create assets, or develop content based on plain instructions. Furthermore, with AI assistants, developers can be guided, automate repetitive tasks, and receive intelligent suggestions. This isn't just hypothetical; innovators are already leveraging custom datasets and models to redefine what's possible.

Digital Twins: Simulating the Physical World

The Omniverse is opening doors to a realm of interconnected digital twins. These aren't just mere digital replicas but can be leveraged using AI, simulation, and real-time 3D to optimize designs even before they're built. This dynamic approach is a leap from static mock-ups, and its applications are vast. Industries like manufacturing, urban planning, architecture, and mining are only scratching the surface of its potential.

Digitizing Physical Products and Experiences

The lines between the digital and physical realms are blurring. With CAD data seamlessly integrating into real-time 3D experiences, opportunities are immense. From interactive product configurators to marketing materials that can be tailored on-the-fly, we're witnessing a new dawn. AR/VR interactions with digital twins further expand the horizons of business potential. Moreover, the concept of owning NFTs associated with virtual products is provocatively blurring the boundaries of ownership between the physical and digital spaces.

At the heart of this revolution is the democratization of creation. AI is not just a tool; it's an extension, an augmentation. With new computing platforms, we're unlocking applications that once lived in the realms of imagination. Yet, as we stand on the precipice of this transformation, it's imperative that we ensure these technologies align ethically with the broader human interests, ensuring an era of responsible and empowered creation.

The Early Days

In the beginning, computer use was the domain of the few. House-sized computers, known as mainframes, were bulky, expensive, and required specialized skills to operate. Then came personal computers, liberating computing power into the hands of individuals. Yet, early PCs still spoke in a cryptic language of keyboard prompts and monochromatic textual displays.

Enter the GUI and Mobile Revolution

The invention of graphical user interfaces (GUIs) transformed the landscape. Computers suddenly became more intuitive, their visual metaphors and point-and-click interactions replicating the physical world. The mobile revolution took this a step further, shrinking computers to pocket size and making interaction as easy as a tap or swipe. Each leap forward simplified computer use and expanded the user base.

The Challenge of Adaptation

Yet, every advancement still asked the user to adapt to the machine. Whether learning DOS commands, understanding the quirks of a word processor, or mastering the swipes and gestures of a smartphone, each upgrade involved a learning curve. This has always been the primary hurdle in the human-computer relationship.

Messaging as a Paradigm

Today, the use of messaging platforms has become ubiquitous, reflecting a shift in how people prefer to communicate. The conversational, asynchronous, and context-rich nature of messaging heralds a new paradigm in human-computer interaction, one that leans towards human preferences rather than machine requirements.

The Limitations of Current Chatbots

Unfortunately, while chatbots have proliferated in areas like customer service, many remain rooted in older interaction paradigms. They're often limited to pre-defined scripts, struggle with context, and fail at handling nuanced conversations, leading to user frustration rather than assistance.

LLMs Breaking the Mold

Here is where large language models (LLMs) like ChatGPT, Bard, and Claude shine. These AI models understand and generate human-like text, enabling them to participate in natural, engaging conversations. By adapting to us instead of the other way around, they represent a significant shift in the human-computer interaction landscape.

Towards a More Human Interaction

The implications of this shift extend far beyond tech. By reducing the need for specialized knowledge or learning new interfaces, LLMs can improve user experiences across a plethora of applications, even heralding a new class of apps that could supplement or replace traditional mobile apps.

Imagine, for instance, booking a flight. Instead of navigating through multiple pages, dropdowns, and fill-in forms on a travel website or app, you could simply tell the LLM your travel plans - "I want to fly to New York from San Francisco on July 15th, returning on July 20th." The LLM could then find the best options for you, asking for additional information as needed, and even complete the booking, including payment, all within a natural, conversational interface.

The same goes for simpler tasks like ordering pizza - you could specify your order, ask for recommendations, make modifications, and confirm your address, all through conversation. This could dramatically simplify the user experience, making technology even more accessible.

Future Perspectives

The next frontier of human-computer interaction has arrived, and it's a game-changer. The paradigm shift towards conversational interfaces powered by advanced large language models is not just a technological evolution, but a revolution that promises to disrupt how we interact with digital platforms. This new era introduces a conversational dimension to digital services that feel as natural as talking to another human.

The evolution towards a conversational user interface is an inevitable and exciting progression. It's an opportunity not just to be a part of the next phase of human-computer interaction, but to spearhead it. Now is the time to invest, innovate, and lead in this promising new world of conversational applications.