top of page

In the age of Artificial Selection: What happens to Venture Capital when code is done before the pizza cools?


Artificial Intelligence has shoved the cost of “trying something” down to the price of a Friday-night takeaway pizza. Two open-weights models, a live data feed, and a Stripe key – that’s now enough to launch a product before the cheese has time to set. The code lands on GitHub, Discord lights up, and by Monday morning the founder has live usage numbers – before a single slide deck reaches a VC inbox.


The startup playbook is being reordered, and Venture Capital is the next chapter under revision. Most likely, this is a once-a-generation expansion of the investment canvas. business.


We now have six-month old startups valued at USD 10 bn (Thinking Machine) and a one-year old hardware-design company (io) being acquired at USD 6.5 bn. These are signals of a brave new world of business. These are signals of a whole new ball game – where AI speeds things up to ludicrous mode and a few extremely capable individuals' prospective futures can be valued to billions.

So, what happens to when these signals have built a new game of business? What happens to best practice of VC investing when AI becomes the status quo? And can AI truly outperform humans?

Impossible to know today. But let's unpack these prompts.



Christopher Lyrhem
Chief Future Officer at Sircular


First off, a couple of days ago, the CFA Institute published a head-to-head test, with six frontier language models versus a team of seasoned equity analysts on SWOT analysis for selection of companies. The machines did more than keep up; they uncovered strategic and governance risks the pros had missed. For example, Gemini Advanced 2.5 produced a board-ready memo in 15 minutes and, with the right prompt engineering, scored 40% deeper than its human counterparts on the study’s specificity-and-depth index.

However, although the bots flagged governance land mines the humans had overlooked, they confidently cited a patent dispute that never existed. The Institute’s conclusive guidance was therefore – treat an LLM like the world’s brightest intern: indispensable for the first draft, never allowed to sign off the final memo.

This tension (brilliance laced with hallucination), is pushing VC firms toward a future where the model does the sweep; while the partner does the veto. Going forward, more funds will likely begin their Monday pipeline calls with an AI-ranked “league table", where the partner debate starts only after silicon has done its cut, freeing human bandwidth for the handful of rule-breakers no algorithm trusts at first glance.

This means that before natural selection of the most intriguing investment opportunities, we'll likely see artificial selection doing its thing.

WEBINAR: AI meets VC - the end of gut feel?

Tomorrow (26th of June) at 3 PM (CET) we're hosting a webinar on this intriguing topic.



Can AI outperform VCs?

A Stanford experiment offers the loudest “yes....but” to date. Researchers trained an “AI analyst” on three decades of public-market data and let it rebalance a model portfolio every quarter. The machine outperformed 93% of professional fund managers and added roughly USD 17m of risk-adjusted alpha each quarter, all without demanding carry.

Yet, the same paper conceded that outperformance narrowed to basically nothing in sectors where data are sparse or ambiguous. Like biotech, frontier hardware, or space. This means that pattern recognition is not disappearing; it is being lifted out of the ordinary and dumped like wholesale on the extraordinary.

The louder the data, the more credible the algorithm. The thinner the signal, the more the market still pays for natural intuition.

Or in other words: AI can win the statistics, while it struggles with the story. Funds that lean exclusively on model output risk underweighting the single anomalous narrative that becomes the vintage-maker.

When Mira Murati recently raised the USD 2bn “mega-seed” for Thinking Machine, no AI-powered spreadsheet wizard would capture the obsessive clarity that convinced investors in the room. And deciding when to pull the board-level handbrake is an ethical calculation, not a gradient-descent optimization.


How will AI shift the profile of tomorrow's breakout founders?

In Business Insider’s latest tiny-team unicorn census that lists AI startups worth USD 1bn or more; none employ more than 50 people. For example, Safe Superintelligence is tracking a USD 32bn valuation with roughly 20 strong, while “vibe-coding” platform Lovable flexes a valuation at USD 1.5bn with less than 40 vibe-employees (and USD 75m in ARR 7 months after launch).

We're now seeing clear signals of valuation/revenue-per-employee reaching previously unimaginable levels. Soon enough, we might have a one-billion company with one founder as the sole employee. Furthermore, in the old venture playbook, the traditional sequence – raise, hire, build, ship – took a year. In 2025 the order is reversed: ship first, raise and hire fewer later. Capital begins as late fuel for something that is proven, rather than early validation discovery.

We're seeing thousands of examples where founders aren't bringing a traditional Minimum Viable Product to the marketplace, enough for a “beta” badge. They're now bringing something thinner and faster, something along the lines of a "Minimum Conversation Loop". Something to continue discussing about. Something to continue develop as we discuss.

Two or three LLM-agents are wired to a live data source, the prompt and the output are public, and users edit both on the fly. Every pull-request, bug report and prompt tweak is visible, measurable, and bankable – long before any revenue line appears. Investors might increasingly care about and lurk in GitHub issues and Discord threads, watching a new metric that never shows up in a traditional data room: feedback velocity. The faster the loop (bug filed at midnight, patch merged before sunrise) the higher the likelihood of success.

This all means that a founder team can do more with less people. They are part engineers, part community managers. Building a prototype that works and getting some customer traction, without capital, is beginning to become an early norm. each other’s services dynamically.


Few-persons VCs can compete with bigger VCs

If startup companies can perform well with few employees, it means that Venture Capital firms can too. With AI, every VC will have the likes of a thousand seasoned investors and researchers at their fingertips. A team of 5 will be able to scale the world's intelligence and compete with current 30-person teams. Theoretically.

They'll have pitch deck analyzers, AI-scoring machines, smart CRM systems, and complex term sheets ready in minutes.

The mega-funds have chosen big league physics instead. They lease private GPU farms, fine-tune internal diligence models on proprietary exit data and sell compute credits back to their Series C companies at cost. The advantage is brute-force certainty, with cost representing maneuverability. Agility or industrial leverage – you can optimize for one, not both, and LPs are beginning to price managers on how deliberately they pick.


Will the best investors in 2030 be ex-operators, quants, or prompt engineers?

Probably all three at once. The partner who can hop from variance math, to founder therapy, to prompt-stack tinkering, without changing tabs – will own the edge. And, just as important, they’ll need to "speak agent” – understanding how autonomous workflows replace human labor costs and how prompt logs replace traditional user-research decks.

In the same way understanding SaaS-dashboards and economics became the norm, tomorrow’s winners will be able to read token burn, retrieval-augmented precision, and how many synthetic teammates a team has deployed in production.


Conclusions – VCs new playbook in the AI-first startup boom

AI is turning every laptop into a mini-factory. Code that once needed a year and a payroll now ships over a weekend, pushed live while the founder’s pizza is still warm. For VCs, that’s not a threat but a flood of chances, with more builders, more experiments, and more surface area to invest:

1 – Speed becomes a super-power: With deal flow exploding, first look and first money matter more than ever. Funds that implement AI-ranked pipelines, deep research, and instant diligence drafts – will outrun slower rivals and lock up outliers before they trend.

2 – Capital becomes a service: Cash alone fades as a differentiator. The winning pitch bundles ready-to-use values like GPU credits, templated legal packs, and a prompt library that slices onboarding from weeks to hours. Large platforms can assemble full tool suites, while focused micro-funds can offer niche expertise and founder-speed decisions. Both models will thrive, but only if they recognize that “money + tools” is the new ticket in.

3 – Lean firms, larger upside: Start-ups show that ten people can build what used to take fifty. VC teams will follow suit. A small partnership armed with strong AI can now match the reach of a 30-person fund. Expect pressure on the 2 & 20 norm, with slimmer bases, flexible carry, and pricing tied to measurable AI leverage rather than headcount.

Final take: General-Purpose Technologies (GPT) have always reshuffled entrepreneurship and VC investing practice. But this one is different, as it runs way faster than past GPTs. The play is to treat AI as both scout and shovel – use it to find diamonds quicker and hand it to founders as part of the cheque. Adapt early and the coming tsunami of creation becomes a once-a-generation expansion of the investment canvas.



Until next time,

Christopher Lyrhem


Chief Future Officer at Sircular







 
 
 

Comments

bottom of page