Kishen Patel

There's more to work than switching tabs

Kishen Patel — Mon, 02 Mar 2026 00:00:00 GMT

Today's AI models are cooperative by default. They're trained to be helpful, and helpfulness means going along with what the user wants. This works fine for drafting emails, summarizing documents, and writing code. But a large share of professional work isn't cooperative at all.

I’m thinking about the roles where people spend most of their time negotiating, bargaining, or managing competing interests. Sales reps working deal terms. Partnership leads navigating revenue shares. Lawyers redlining contracts. Finance teams haggling over vendor pricing. Executives fighting for headcount, defending budgets, or making tradeoff calls on scope, timelines, and resources. For many roles, this is the job.

Consumers face fewer bargaining scenarios, but the ones that exist tend to be the biggest financial decisions of their lives: buying a car, negotiating a home purchase, scoping a renovation, disputing medical bills.

The demand for AI that can operate in these settings is obvious. For these use cases, today, we're in a copilot phase where an agent might guide you through the process and compile information, but an agent that can negotiate your lease renewal or push back on a vendor's pricing without your input would be enormously valuable. It feels obvious that we'll get there, but the moment you optimize a model to win rather than to please, you change the alignment problem entirely. And "surprise," the moment when the other side does something the agent didn't expect, is where that new alignment problem shows up first.

Surprise is a key part of reasoning

People sometimes claim that LLMs can't be surprised. There's a version of this that's technically true and a version that misses the point.

On the technical side, transformers have architectural limitations around being "caught off guard." They process every token with the same machinery, whether the input is ordinary or bizarre. They have no interrupt signal, no "oh wait" reflex. And current post-training approaches like RLHF (the process used to make models helpful) push models toward predictable outputs, which makes them look even less capable of registering surprise.

Yet these models traffic in probabilities by design. They can assign high "surprisal," an information-theoretic measure of how unlikely a given token is, to low-probability events. The model "knows" when something unlikely has happened. The question is what it does with that information.

Think about what surprise looks like for an agent negotiating a vendor contract. The vendor suddenly drops their price by 30% with no prompting. Or they introduce a new requirement late in the process that changes the deal structure. Or they go silent for a week after you expected a counteroffer. Each of these is a moment where the agent's assumptions about the other side no longer match what's happening. That mismatch is what we mean by surprise in this context: the gap between what the agent expected and what it actually observed.

In game theory, that gap is called a belief update, and it carries strategic information. When your opponent makes an unexpected move, it tells you something about their strategy, their model of you, how much risk they're willing to take. Good strategic play means revising your model of the other side in response to that signal. A skilled human negotiator uses surprise productively. They notice the unexpected move, ask themselves why it happened, and adjust. The question is whether an AI agent can do the same.

This is where the gap between cooperative and competitive AI is most visible. Strategic reasoning requires game-theoretic thinking, where the model accounts for other players' actions instead of optimizing its own output in isolation. You need max-min optimization, which means maximizing your reward while assuming others are trying to make you worse off. And you need theory of mind: the ability to anticipate what opponents are thinking and will do. Current models can't do this well. They optimize for the next token given what they've seen, without modeling how an opponent will respond to their move.

The training data is part of the problem. Supervised fine-tuning uses loss functions designed for single-agent input-output mapping. The data typically contains one player's actions, not the interplay between two adversaries. The AlphaGo lineage offers a partial blueprint: self-play, where a model plays against previous versions of itself, generates the kind of adversarial training data that strategic reasoning demands. But the challenge with language, as Andrej Karpathy has noted, is that there's no equivalent of a perfect game rule. Go has a clear win condition and a perfect world model. Negotiation doesn't. The reward signal is noisy, delayed, and often subjective.

Recent research shows these limitations in practice. A NeurIPS 2025 paper tested 22 LLMs through behavioral game theory. Models like GPT-o1 default to minimax reasoning. This works well in purely competitive (zero-sum) games but becomes too cautious in settings with mixed incentives where cooperation is sometimes optimal. GPT-o3-mini showed more flexibility, attempting to infer the opponent's likely move rather than assuming worst-case play. In other words, the models that handled surprise better (by updating beliefs about the opponent rather than retreating to a fixed strategy) performed better in realistic competitive settings.

Alignment Tensions

So we have a valuable application (competitive agents) and a specific technical gap (handling surprise strategically). The harder question is what happens when we start closing that gap. Training models to be competitive introduces a set of alignment tensions that don't exist in cooperative AI.

What counts as "aligned competitiveness"? Optimal play in negotiation can include bluffing, strategic omission, even threats. Most people would call those behaviors misaligned in a customer service chatbot. A negotiation agent can't be aligned to a single abstract virtue like honesty. It needs a spec that says "win, but only within these constraints," and those constraints have to be specific enough to audit. A large-scale AI negotiation competition running 180,000+ negotiations found that warmth produced the best outcomes. Exploitation-based strategies like prompt injection ranked in the bottom 12th percentile for counterpart satisfaction, and this lopsidedness hurt their overall performance. Even in competitive settings, choosing the right alignment constraints can mean choosing the best strategy.

Competitive training creates new incentives around surprise. If the training reward is "win," agents will figure out that surprise is something to manage strategically. They might learn to hide uncertainty, feign confidence, or selectively disclose only the uncertainties that don't cost them. We already have evidence that models develop this kind of strategic behavior on their own. In Anthropic's alignment faking research from December 2024, Claude 3 Opus was told it was being retrained to comply with harmful queries. The model began complying with harmful requests from monitored users 14% of the time while refusing the same requests from unmonitored users. Its own reasoning made it clear: it was complying strategically to preserve its preferred behavior through training. Nobody trained it to do this. Now imagine that same capacity for strategic self-presentation in an agent whose explicit objective is winning a negotiation. It will quickly discover that appearing transparent is different from being transparent, and that the optimal amount of visible surprise is whatever keeps oversight satisfied while preserving strategic advantage. The competitive training objective would amplify exactly the behavior that alignment faking showed us models can already do.

Surprise is where evaluation breaks first. Benchmarks are built from tidy, on-distribution interactions where it's easier to look good. Competitive agents need evaluation that looks and feels like deployment: messy opponents, shifting tactics, rare tail risks, long time horizons, and real stakes. And there's a deeper problem. If agents can infer they're being tested, they'll perform cautiously, comply, and "behave well," then switch modes in real interactions. "Surprise" effectively becomes a proxy for "this is the part of the distribution where someone is watching." Output monitoring has the same problem. A negotiator optimized for outcome will notice what the monitor flags and shape its behavior accordingly. It might shift from outright deception to more ambiguous language, or move manipulation into parts of the interaction that the monitor doesn't model well. The agent has an incentive to stay just inside the boundary of what looks acceptable.

The Path Forward

The sales rep who needs an agent to negotiate vendor contracts, the consumer disputing a medical bill, the executive defending a budget: they all need an agent that can win on their behalf. From my point of view it’s inevitable that there will be a large market here given the volume and value of tasks, but building these agents into production systems will rest on how well the agent handles surprise.

Getting there requires progress on three fronts.

Training needs adversarial data that captures the back-and-forth of real competition, not just single-player input-output pairs.
Evaluation needs to test agents in conditions that actually surprise them, not sanitized benchmarks they can pattern-match against.
Monitoring needs to distinguish between an agent that is genuinely uncertain and one that has learned to perform uncertainty because it's useful.

We don't have that agent yet. But the demand is here, and the companies that figure out how to make competitive agents handle surprise well, rather than just hide it well, will be the ones that build agents people actually trust with high-stakes decisions.

As always, if you’re thinking about, tinkering or already building in this space, please reach out. I’d love to hear your perspective.

NeurIPS

Kishen Patel — Tue, 09 Dec 2025 00:00:00 GMT

NeurIPS – the world’s largest AI and machine learning conference – was last week in San Diego. This year's conference felt particularly significant: over 26,000 attendees packed the convention center and the crowd was as much (if not more) investors and industry as it was researchers, reflecting the intense focus on AI these days.

At a meta-level, the field feels like it is at an inflection point: many of the scaling assumptions of the past few years feel exhausted or at least delayed by what hardware can actually deliver. I lost count of how many conversations circled back to this same sentiment. The natural response, it seemed, was a pragmatic pivot: instead of reaching for more compute or more data, everyone was asking how to squeeze more from what we already have.

If compute-scaling hits a wall, how do we use hardware more efficiently? If we've scraped the internet dry, how do we get more learning turns from the same data? If algorithmic advances are slowing, what new architectures will actually move the needle? These questions framed the conference and pointed toward a future where resourcefulness matters more than raw scale.

Of the research themes that kept surfacing, three stood out:

Rethinking Model and System Evaluation

There's an evaluation crisis brewing. Most benchmarks shaping the public narrative around LLMs were created years ago for models a fraction of the size we have today. They suffer from two fundamental problems.

First, scale mismatch: models trained on trillions of tokens are evaluated on test sets of just thousands of examples. The results are statistically fragile at best. Second, contamination is everywhere. Public benchmarks have leaked into training corpora, inflating scores that don't reflect genuine generalization. Experiments showed substantial inflation for open-weight models, and speakers warned that contamination propagates subtly. Even synthetic data from contaminated models can poison downstream systems.

Then there's the science of measurement itself. One tutorial argued that LLM evaluation lacks empirical rigor. Metrics get reported without confidence intervals, meaning apparent score differences might be statistical noise. The field is shifting toward grounded, task-specific evaluations that measure performance on economically meaningful work. Scale AI's Remote Labor Index was a good example where they test agents on actual freelance marketplace tasks. It was clear that model value isn't determined by marginal leaderboard improvements, but by performance on tasks that matter to end users and businesses.

RL and the Rise of Continual Agentic Systems

Continual learning emerged as a clear priority. Agents should learn from experience over time (and meta-learn how to generalize) rather than through one-off training runs. In his invited talk, Rich Sutton argued that robust intelligence requires agents with world models, planning abilities, and continuous learning capacity. His proposed OaK Architecture was one approach: a model-based RL system where every component including the learning rates themselves adapts during online training.

Other sessions highlighted efforts to connect LLMs and vision models to embodied intelligence. Tesla's robotics panel detailed their work training large-scale multimodal models for end-to-end "pixels-to-actuation" robot control. Autonomous vehicle discussions emphasized open challenges in scaling to full self-driving: architecture choices, balancing imitation learning versus RL, and the infrastructure required for large-scale simulation.

What's missing and creating bottlenecks? Infrastructure. Beyond training environments, the field needs better simulators, reproducible safety benchmarks, and standardized environment-model integration.

Scaling Architecture and Efficiency

Targeted architectural changes can yield significant gains. One award-winning paper showed that adding a head-specific gating mechanism to standard attention layers notably improves performance. This simple sigmoid "gate" introduces sparsity and non-linearity, enabling more stable training and better long-context handling. The technique has already been adopted in models like Qwen-3 and helps mitigate the "attention sink" effect (where models overfocus on a small subset of tokens). The broader trend is active experimentation with modified attention, memory mechanisms, and alternatives to the standard Transformer architecture. Google debuted their Titans architecture and MIRAS framework as one such alternative.

On the efficiency front, Microsoft researchers introduced orthonormal updates (optimizers like Muon and Dion) as potential successors to AdamW for large-scale training. These improve convergence in deep or distributed settings. A dedicated session on replacing Adam signals how seriously the field takes any speed or stability edge. On the inference side, a tutorial on test-time compute covered caching, retrieval, and on-demand computation techniques. Open-source projects like vLLM and sglang were also top of mind for many.

Where This Leaves Us

The convergent theme is maturation. We're shifting from scaling what works to understanding why it works and then making it work better. The evaluation crisis forces us to confront what we actually want from these systems as opposed to what we can measure. The push toward continual agents recognizes that intelligence is fundamentally about adaptation, not pattern matching. And the architectural innovations suggest that even our most fundamental building blocks are still open to reinvention.

NeurIPS 2025 felt like a necessary recalibration. The community is homing in on a nuanced set of principles: efficiency, robustness, and genuine capability over leaderboard points.

The path forward looks less like a straight line upward and more like a careful, deliberate hill climb.

Bubble Machines Make Bubbles

Kishen Patel — Tue, 28 Oct 2025 00:00:00 GMT

Every other conversation I’m having these days is about whether or not we are in an AI bubble. Coatue has argued we’re in a durable AI supercycle, not a bubble, while Derek Thompson has laid out plausible paths for an AI bubble to burst. I’m not interested in sharing yet another take on Bubble Watch 2025, but I want to offer some observations on market cycles in early-stage venture markets.

Venture capital is unusually but structurally bubble-prone. In public markets, by contrast, every trade nets a buyer and a seller with (roughly) opposite beliefs. The optimist goes long; the pessimist sells or shorts. In early-stage venture, the “seller” is typically a founder issuing new primary shares, who remains massively long the company, since they often are the majority owner and the company is their livelihood. Of course, the “buyer” is a VC who is also long.

There’s no natural short, and skepticism can’t easily be expressed in price. Classical finance predicts that when disagreement is high, yet short-selling is constrained, prices reflect the optimists’ views, not a balanced consensus. While in the long run every market is a weighing machine, early-stage venture in the short run is a voting machine with only one button.

The result is a market with a unique capacity for bubbles: belief meets belief, capital is plentiful, and price discovery is slow. When fund inflows accelerate and “money chases deals,” they raise deal prices as Gompers & Lerner documented two decades ago. But we don’t need to go that far back: we saw a modern replay in 2021 as step-ups and median valuations hit records. When the tide turned, down rounds and valuation resets followed as nearly one in five priced rounds were down rounds by late 2023.

This market structure changes the craft of investing. The game is usually framed as “find the winning needle in the needlestack”, and selection matters because venture returns are power-law distributed. But that is only a partial truth. The lesson of the power law isn’t that “selection is all that matters,” or that being in the right names alone is enough. Instead, it’s that selection along with sizing (the amount of exposure taken) and timing (the terms and moment selected) are the only reliable tools an investor has to make the power law work for the fund and its LPs rather than just for the company. Yes, a single investment will often return the fund and more, but that is exactly why overpaying for even the right company, or owning too little of it, sinks fund-level outcomes.

In euphoric periods, investors tend to neglect the less glamorous half of the job. They stretch on price (timing), accept thinner ownership (sizing), or front-load enormous initial checks (also sizing) because “this is the one.” Deals get papered in lightning-fast processes that prioritize speed over price discovery.

For my founder friends reading, it’s worth noting that this is all founder-friendly and great for getting capital raised, but it can make it hard to know who the true believers are since it can be easy to misread speed as conviction rather than FOMO. But it comes to light when the company inevitably finds itself in a different market environment and those tourist investors find themselves with thin ownership in an overvalued company. What ensues usually isn’t pretty.

At the ecosystem level, funding many at-bats from many founders is good, especially when less euphoric times might not have funded those startups. It’s a key part of our innovation-driven economy. But at the fund level, paying any price for those at-bats is not good. An investor’s job is to fund innovation and deliver returns — in other words, survive.

The paradox is that when everyone focuses on picking, the marginal edge often lives in sizing and timing. The best compound an entry edge into a fund-level edge through discipline. They demand adequate ownership for the risk taken and cautiously allow price to affect their appetite even for beloved and most obvious companies. They also use pacing, reserves, and pro-rata selectively. This can feel like moving against the cycle instead of with it, but it forces real conviction.

None of this, though, is an argument for stinginess or market-timing heroics. Meeting founders where the market is will still be required. Not doing a great deal or sitting out a super-cycle because of price is a risk unto itself.

But this is a reminder that venture’s structure of optimists transacting with optimists, little shorting, and capital that arrives in bursts makes it unusually vulnerable to narrative overshoots. When belief is abundant, selection converges (most people agree on the same handful of names), so sizing and timing become the scarce behaviors that actually differentiate returns.

Put differently: in cold markets you win by being willing to believe and in hot markets you win by being willing to budget and pace that belief.

The Public Option

Kishen Patel — Fri, 20 Jun 2025 00:00:00 GMT

After a long pause, tech IPOs are finally showing signs of life again, with recent listings from CoreWeave, MNTN, and Chime. It feels like the perfect moment to revisit a critical question: Should founders and VCs really care if their companies ever go public?

There’s a common argument out there that venture capital should chase returns wherever they come from—private secondaries, acquisitions, buyouts, or IPOs. Founders often share this perspective; after all, either route typically leads to life-changing wealth. Plus, running a public company isn't exactly a walk in the park. Reporting, earnings calls, quarterly pressures—who wouldn’t want to skip that headache? Maybe that's why we've seen fewer companies going public over the past couple of decades.

But here's my take: if you're an ambitious founder or venture investor, going public should still be your ultimate goal. Why? Because the truly generational businesses driving lasting impact are almost always public.

The reason boils down to something deceptively simple: public companies have a lower cost of capital. Public markets dwarf private equity markets ($88 trillion vs. $11 trillion globally), making them deeper, more liquid, and cheaper places to raise funds. Simply put, more investors competing means better terms, lower costs, and a bigger strategic toolkit for the companies that list.

Having cheaper capital isn't just a nice bonus—it’s a powerful strategic advantage. Public companies can afford mistakes, survive downturns, and aggressively chase big, ambitious projects that would bury privately funded competitors. Over time, this lower capital cost creates a compounding competitive advantage, turning good companies into great, enduring ones.

It's easy to overlook this in the early stages, but growth almost always requires fresh capital. Growing solely from revenue can work, but it’s often painfully slow, an unaffordable luxury in hyper-competitive sectors like tech. Just look at Meta: even as a profitable giant, it raised $10.5 billion in debt last year to fund ambitious AI initiatives. Public companies simply have unmatched flexibility and access to capital for strategic acquisitions and bold moves.

There are two main knocks against going public: short-termism hurting innovation and burdensome reporting. On innovation, academic evidence is mixed, but plenty of studies suggest being public doesn't necessarily stifle creativity. In fact, some suggest it can even boost innovation by expanding resources. And honestly, founder-led tech companies tend to ignore short-term market noise anyway. These founders spent years building their vision; a bad quarter isn’t likely to shake their conviction.

As for avoiding reporting hassles by staying private: sure, that’s appealing today. But those private companies are essentially piggybacking on the transparency that public companies provide. Investors in late-stage private rounds rely heavily on publicly listed companies as benchmarks. If too many great businesses stay private, transparency shrinks, benchmarks degrade, and eventually, capital costs rise for everyone.

Bottom line: going public isn't easy, but it’s still the best way to build a lasting, impactful business.

But why should investors care whether it’s an acquisition or IPO if they're selling either way? Public market investors typically pay premium valuations at IPOs because they price in a company's long-term, generational potential. These premiums often translate into one or two additional turns of revenue, or about 22% more than what acquisition counterparts would pay. Moreover, backing companies that grow into generational giants has powerful secondary effects for venture investors. It builds reputation, boosts deal flow, and enhances the fund’s ability to attract future capital and talent. Even though VCs might not hold shares through decades of public growth, investing in future public winners significantly raises their long-term success and influence.

Ultimately, the best founders view their companies as their life's work and their lasting mark on the world. They’re deeply passionate about solving meaningful market problems and continuing to tackle new challenges as their company grows. That’s true founder-market fit. To keep solving meaningful problems at scale, over decades, these founders typically need the scope and resources of a public company. Palmer Luckey recently made exactly this point regarding a potential Anduril IPO: “We are running this company to be the shape of a publicly traded company. There isn’t really a path for a company like Anduril winning things like an F-35 joint strike fighter contract as a private company.”

At the end of the day, venture investors’ core job is simple: identify and back these ambitious founders and companies that are destined for public markets. Doing so positions investors to capture exceptional returns and sets the stage for genuine, lasting impact.

404: Human Not Found

Kishen Patel — Thu, 27 Mar 2025 00:00:00 GMT

I often get asked about how I expect AI to reshape our lives in the coming decades. While I don’t know anything with certainty, I'm seeing small breadcrumbs of the massive transformation ahead.

One of the most fascinating shifts is in the window you’re staring at right now: it's who — or what — the internet is built for and the way we interact with it (via the browser). The internet was originally created to connect computers and let humans share information in standardized ways (namely HTTP). These connections turned out to be revolutionary — they created new business models and took down old ones. Now we're witnessing something equally profound: the transition from a human-first to a bot-first internet. The full implications are not immediately clear, but they could be equally impactful to society.

The story starts with web crawlers, the internet's first bots, that mapped out every corner of the web to make information searchable. Website admins quickly realized they needed some control over this and came up with robots.txt — a simple way to tell bots which parts of their sites were off-limits. Adherence was voluntary: while reputable players like search engines mostly played by these rules, others didn't bother.

As search made the web more accessible and PageRank morphed into SEO, optimizing engagement through UIs took center stage and web testing frameworks grew alongside. A pivotal moment was in 2004 when Jason Huggins created JavaScriptTestRunner, which was open-sourced as Selenium Core and injected JavaScript into pages to simulate user actions. Soon after, Simon Stewart introduced WebDriver, which natively controlled the browser and provided a higher-level API. Later, the two combined into Selenium WebDriver, which became the de facto standard for browser automation testing until 2020, when Microsoft open-sourced Playwright.

As web automation tooling grew in testing use cases across a growing landscape of websites and browsers, it supported scale in another market — malicious bots spamming forums, flooding inboxes, and harvesting data. These days, bots are behind 77% of online security incidents. We've fought back with reCAPTCHAs, behavior tracking, and rate limits. Most websites now treat non-human traffic as guilty until proven innocent — and if you’ve seen spaces like X lately, you might agree that they've had good reason to.

But LLMs are eating away at this regime. These new AI-powered bots (okay fine, agents!) can navigate the web with flexible goals, almost like humans. I see three avenues LLMs can take for web-based automation: multimodal point-and-click automation, navigating the DOM through LLMs, and making direct API calls.

Multimodal point-and-click. The most intuitive approach to web automation would be to just replicate how humans interact with the web (look at a page, look for an action that matches my objective and click!). Systems like Operator and Computer use take screenshots that are sent to a model that instructs the next action. In its current form, I don’t have much faith that this will be the winning solution for a few reasons. First, processing images is compute-intensive and is compounded by the fact that we require a headful browser to render pages. Second, we’re not guaranteed to get the full set of actions when the page is loaded. How many of the actions sent to an LLM will just be scroll down… okay now… scroll up? This is of course great for generating tokens for your foundation model business, but not so great for scalable application-layer use cases! The bright spot here is that it implies existing UI patterns will suffice for AI, and there is hope if we can train models to read and interact in a compressed image space to save compute.

Navigate the DOM with LLMs. Another option is to use LLM-based code generation to write Puppeteer or Playwright scripts to accomplish tasks. With services like Browserbase, Steel, and Browser Use, I can supply a flexible goal to an LLM, which will look at the site’s DOM and generate the script on the fly. I like this approach for two reasons. First, brittleness is why most automations fail — the button moved or the URL path changed. Even the best scripting frameworks will require too precise a specification and will result in hours of maintenance when things break. Natural language allows richer flexibility and allows for orientation around goals rather than process (the what rather than the how). Second, these can be run in headless browser instances suitable for larger-scale cloud deployments and generally work well with the existing automation infrastructure.

Make direct API calls. A final way to automate the web is by skipping the browser altogether and directly calling APIs that generate the information on a page. I’m sure someone somewhere has already said “APIs are eating the world”, but AI is enabling API integrations at an accelerating pace. Sites with APIs that don’t want to spend time developing and maintaining SDKs can use AI tools like Stainless to auto-generate them based on an OpenAPI spec. And when sites don’t have public APIs, services like Candle and Integuru trace requests and use AI to reverse engineer internal APIs into SDKs.

As LLMs further influence how we interact with the internet, there are two interesting implications.

The litmus test for unwanted traffic will change. It seems likely that many of us will be okay with a bot taking actions on our behalf on the web. But for that to happen, we will need to have a way to both authenticate and authorize these bots, especially as their traffic becomes indistinguishable from human traffic. I don’t expect this to be a passive protocol like robots.txt, but it could be a sort of accreditation for legitimate traffic — like a license plate that interacts with some combination of cars (browser automation platforms), toll booths (web auth like Okta), and roads (network services like Cloudflare).

Websites will look different as they become dually optimized for human and bot use. One of the reasons I struggle with the multimodal point-and-click approach is that it limits models to conform to how humans traverse the internet. It feels like the “Waymo stage” of web automation, where our human-readable UIs are steering wheels that surely will be replaced by more information-dense bot-first interfaces. If I take each web automation approach above to its extreme, it leads to roughly the same place — LLMs that parse a large block of dynamic information and serve that information with a UI back to users. In other words, the models themselves eventually become the browsers we interact through.

What makes this moment so thrilling is that web automation — once the domain of QA testers and spammers — is evolving into something magical. With AI at the helm, we’re not just scripting clicks; we’re teaching software to problem-solve, navigate, and act on our behalf. The browser, once a pane of glass between us and the internet, may soon become a conduit for intelligent agents that understand our goals and reshape how we experience the web entirely. It’s a small slice of a much bigger shift — one where AI doesn’t just augment our lives, but rewires the very interfaces we use to live them.

Oh so you're technical?

Kishen Patel — Wed, 19 Feb 2025 00:00:00 GMT

The quality of being technical dominates the early-stage ecosystems — both among founders and VCs — in a way I’ve never understood.

I remember mentioning to a founder that I had been interested in machine learning and computer vision in undergrad when they interrupted to say, “Oh so you’re technical?” I paused and wasn’t really sure how to answer. Sure, I had studied these models and implemented algorithms for homework assignments but that seemed far from calling myself technical. To me, the technical expertise was with the researchers inventing these models and algorithms, from which I was a world away as an undergrad.

But I presume this founder meant technical in a different, but much more common parlance. Many use technical in a weird have and have-not sense — you either are technical, typically based on majoring in Computer Science in undergrad, or you are not (sorry for you!).

It has become a moniker to signal being in some in-group of the knowledgeable few. “Don’t worry, I’m a technical VC so I know what I’m talking about.” Oh really? You somehow have cutting edge knowledge of every forefront of technology you will encounter in a job whose entire premise is investing in a future we can’t know today?

It’s helpful to crack open the dictionary for this one.

technical. adj. having special and usually practical knowledge especially of a mechanical or scientific subject.

I would point out here that technical means knowledge within a domain — it is not a generalizable quality of a person. You might be a technical expert on K8s deployments or on mobile front-ends, but you are not just technical. Additionally, this knowledge exists on a continuum rather than in a fixed state, and this is especially true of technology which evolves every day. Others may have more knowledge in an area than you and if you do not continue to develop your knowledge, it may no longer be considered particularly specialized.

So when I say I’m investing in technical founders, I quite literally mean these qualities: (a) deep specialized knowledge of their problem space and (b) a desire to remain at the forefront with that knowledge. Notice that none of this inherently requires a CS degree: it may, depending on the problem space, but just claiming to be technical on that fact alone is not convincing enough.

But in terms of how this founder was using the term technical for me as an investor - as a litmus test on some basic knowledge of their space - there is still merit to this and so I propose a new way to understand what technical means. Technical in this sense means “do you give a damn about what I’m interested in?” Today, knowledge is all around us and it is easier than ever to get a baseline in any space. Documentation abounds and can be consumed via LLMs in a variety of formats, whether through chat, summarization, personalized podcasts, Deep Research and voice-based conversations — the list just continues to grow. Getting to the 70th percentile of knowledge in a space is doable if you are interested enough to spend the time. So to be technical just means a willingness to spend time to truly learn how a novel technology works at a fundamental level.

It is the quality of acting on the curiosity that many claim to have as investors.

So the good news is anyone can be technical, not just CS majors! But the bad news is that we have to work for it and constantly upgrade our knowledge. Fortunately, this is what lovers of technology — whether founders or investors — just do anyways.