A new category of company is emerging in Silicon Valley. Not AI-enabled. Not AI-assisted. Not AI copilots. But companies designed around a far more radical premise: that orchestrated AI agents can run large portions of a business autonomously — and that human headcount is now a strategic choice rather than an operational necessity.
The clearest example right now is Polsia. Its positioning is unsentimental: "AI that runs your company while you sleep." And unlike most AI startups, Polsia is not selling automation software. It is testing a completely different organisational model — one in which the company itself is the product, and the platform operates a fund of autonomous businesses on behalf of thousands of customer-founders.
Polsia matters not because every claim will hold at enterprise scale. It matters because of what it makes thinkable. And as we will show, it is no longer alone.
What Polsia actually is.
Founded by Ben Cera — a former CloudKitchens operator under Travis Kalanick — Polsia announced a $30M Series A at a $250M post-money valuation in May 2026, led by Sound Ventures with True Ventures, Adjacent, Tekton, Offline, Vaynerfund and others participating. The fundraising was itself a demonstration: Cera says the AI handled the data room, investor briefings and diligence, while he joined only the final calls.
One founder. Zero employees. $10M ARR in five months.
The platform charges $49 per month plus a 20% revenue share — an unusually aggressive economic structure that treats AI infrastructure as a partner taking a cut of the business it builds, rather than a tool taking a flat SaaS fee. The agents handle market research, code, advertising, customer support, cold outreach and ongoing operations. Cera has even floated giving the AI 10% equity in the parent entity.
Strip away the founder's showmanship and the operating model is what matters. Historically, companies scaled through headcount, departments, middle management, offshore labour, outsourced operations and software systems that supported humans. Polsia inverts the entire stack:
This is closer to cloud infrastructure for business operations than to traditional SaaS. The product is not a tool. The product is the company.
Polsia is not alone.
If Polsia were a one-off, it could be dismissed as an artefact of frothy capital markets. It is not. A small cohort of companies has now demonstrated the no-human or near-no-human operating model at materially different scales — and across materially different sectors.
Three cases. Three sectors. One operating premise.
Medvi
"A one-person GLP-1 business on a $1.8B trajectory."- $401M Year-1 revenue From a September 2024 launch with no employees
- 250,000 customers ~12 AI tools handling acquisition and operations
- 16.2% net margin Versus Hims & Hers at 5.5% with 2,442 employees
- $20k founding capital Matthew Gallagher, solo founder
Base44
"Built solo in six months. Sold to Wix for $80M cash."- $80M acquisition price Plus a further $90M in earn-outs
- 250,000 users at exit No co-founder. No seed round. No team Slack.
- $189k profit · final solo month Maor Shlomo, solo operator
- 6 months launch-to-exit Profitable. Solo. Sold.
Pieter Levels Portfolio
"PhotoAI, NomadList, RemoteOK — one operator, three products, a decade of compounding."- $3–5M annual revenue Across the portfolio
- 3 live products All operated by one person
- 0 employees Decade-long proof of the solo-operator model
- ~80% AI execution share Step-change in per-product velocity since the AI wave
Anthropic's Dario Amodei placed 70–80% odds at the May 2025 Code with Claude conference that the first solo-founded, billion-dollar company would emerge by end of 2026. Sam Altman has been running a betting pool on the same question in a private group chat with fellow tech CEOs. As of May 2026, Medvi sits at the front of the field. Two-person AI companies have already crossed $1B in revenue.
Why Fortune 500 boards should care.
The wrong question for a Fortune 500 board is "can AI replace our employees tomorrow?" That framing locks the conversation into headcount maths and union politics, and it misses the strategic point entirely.
The right question is this: entire new business units, departments and service lines can now be designed AI-first from day one — without retrofitting anything.
The enterprises that learn to launch greenfield AI-native units alongside their existing human organisations will compound a structural cost and speed advantage that is mathematically impossible to match through traditional efficiency programmes.
Six no-human departments a Fortune 500 could launch in 2026.
These are not speculative. Every component below is technically feasible today with the current generation of frontier models and agent frameworks. The constraint is organisational, not technical.
Six departments to launch — and what each replaces.
Each can be ring-fenced under a single human director, run alongside the existing function, and learn fast without political resistance.
AI Revenue Operations — continuous, not Monday-morning
AI Market Intelligence — briefing every morning, not deck every quarter
AI Procurement Operations — continuous loop, not quarterly cycle
AI Customer Expansion — the team that actually knows the 200 accounts to call
AI Compliance & Risk Monitoring — live telemetry, not periodic audit
AI Outbound & Demand Generation — governance, not feasibility
Each of these can be launched as a ring-fenced unit reporting to a single human director, with a small oversight team, and run alongside — not in replacement of — the existing function. This is the path of least political resistance and the highest velocity of learning.
The 100× productivity question.
Ratios of revenue per employee have become the most interesting single metric in software. Midjourney generates roughly $18M per employee. Medvi crossed $401M with one person. Compare that to the median Fortune 500 software business, which sits between $300k and $600k per employee. We are looking at productivity differentials of 30× to 1,000×, depending on the benchmark.
The implication for board-level capital allocation is severe. A traditional F500 unit might add $50M of revenue for $40M of fully loaded cost. An AI-native unit aiming at the same revenue might cost $2–4M to operate. The capital that previously funded headcount becomes available for distribution, acquisition, or further AI-native unit creation.
This is also the mechanism by which the productivity differential gets locked in. Once a competitor has redeployed three to five units onto AI-native operating models, the laggard cannot catch up through hiring. The only catch-up move is to build their own AI-native units — at which point the gap compounds further while transformation budgets get spent on change management rather than capability.
The caveats worth saying out loud.
A serious editorial position requires acknowledging where the model breaks. Polsia itself is a useful case study in the gap between promise and execution.
Polsia's customer reviews on Trustpilot average 2.1 out of 5, with users reporting tasks marked complete that never actually deployed, and credits burned on failed agent runs. Cera himself has been transparent about most outputs being low quality, arguing that quality is improving week by week. The investor pricing — $250M post-money — assumes that reliability curve will resolve. It might not. For a Fortune 500 board, the lesson is not "don't do this." The lesson is that AI-native units need radically better governance, observability and human-in-the-loop checkpoints than venture-backed startups are currently building. The autonomous loop needs an audit layer. This is a procurement and architecture problem, not an excuse to defer.
A solo founder generating "AI slop" content from 20 X accounts is an inconvenience. A Fortune 500 brand doing the same is a Wall Street Journal article. The governance bar for AI-native units inside large enterprises is materially higher — particularly in regulated industries.
AI-native units do not eliminate the need for talent. They concentrate it. The director running a six-agent revenue operations unit needs to be sharper than the VP running 30 humans, because they are designing the system rather than managing it. Fortune 500s that view this as a cost-cutting exercise will under-invest in the orchestrator layer and the units will under-perform their potential.
The operating model beneath the headlines.
The most important shift is not the absence of humans. It is the shift from human shifts to AI shifts. Traditional companies operate around office hours, departments, human scheduling and geographic staffing. AI-native firms operate on continuous operational cycles, autonomous execution loops, on-demand agent teams and elastic intelligence capacity.
Polsia reportedly runs nightly operational cycles where the AI reviews each customer business, identifies priorities, executes work, and delivers updates automatically. That sounds small. Conceptually it is enormous: businesses begin to operate through continuous intelligence systems rather than static human workflows.
Previous automation waves optimised repetitive tasks. This wave targets decision-making, coordination, execution, analysis, operations and communication. The unit being automated has moved up the stack from task to function.
What this means for the GTM layer.
The GTM function is the most exposed of any inside the enterprise, because it is the most measurable, the most software-touchable, and the one where execution lag costs the most. Pipeline review cycles, account research, outreach personalisation, intent monitoring, sales enablement — these are exactly the workflows that AI-native units now do better than human teams.
The honest forecast is this: within 36 months, the highest-performing Fortune 500 revenue organisations will not be the ones with the largest sales floors. They will be the ones running a small, sharp orchestrator team on top of an AI-native demand engine, an AI-native customer expansion unit, and an AI-native market intelligence loop — all reporting into the CRO.
This is the GTM operating model we have been mapping in this Review for the past several issues. Polsia is not the model. But Polsia is the proof that the model works.
The board's takeaway
Polsia may ultimately become a giant company, a niche experiment, or simply an early proof of concept. Medvi may stumble. Base44 may turn out to be a one-off. None of that matters very much.
What matters is that the autonomous enterprise has now been demonstrated empirically — not in a McKinsey forecast deck, but in revenue, customers and acquisitions. A company no longer needs large human teams to operate meaningful business infrastructure. That single idea reshapes enterprise software, consulting, outsourcing, operations, labour markets, and the structure of the modern corporation itself.
The Fortune 500 firms that learn from this shift earliest — by launching their first AI-native units in 2026, not 2028 — will define the next era of global business. The rest will be writing the case study on themselves.
This is Part I of the Autonomous Enterprise Series. Part II will examine the orchestrator layer — the small, high-judgement teams that sit between the CEO and the agent fleet — and what it takes to build one.