The 20-Person Company Is Disappearing — Which Company Form Fits the AI Era?

Last month I wrote about the Stack Ownership Spectrum — which layer of the tech stack your product should own in the AI era. The most common response wasn’t about product forms. It was: “What about the company itself? Solo or team?”

This question is more personal and more anxiety-inducing than product form.

I run a one-person company. Every day I collaborate with a fleet of AI agents — they mine demand, write code, generate content, monitor data. Honestly, most days I feel productive. But I also know the limits: enterprise clients won’t trust a solo operator with critical systems, I can’t do 24/7 support, and compliance isn’t something AI can handle.

So I started thinking seriously about this. Not “is a one-person company good” — that’s a 2024 debate. But: what company forms actually exist in the AI era, and what can each one do and not do?

After mapping it out, I found the most dangerous position isn’t any single form — it’s being stuck in the middle at 20-200 people. Before explaining why, let’s look at the configurations.

Not Everyone Will Become a Solo Company

Start with a data point.

In 2024, a good SaaS company generated about $300K revenue per employee. By 2026, AI-first companies are rewriting that number — Midjourney reached hundreds of millions in revenue with fewer than 20 people, and Cursor’s parent company Anysphere hit $100M ARR with roughly 20 employees. $100M ARR with under 100 employees is becoming a new category.

But $3M per employee doesn’t mean “fewer is always better.” Instagram had 13 people when acquired by Facebook, WhatsApp had 55 — both “small,” but different configurations for different problems.

The question isn’t “should I use AI to cut headcount.” Every AI-era company is doing that. The real question is: which configuration should you choose?

6 Company Forms

I’ve mapped AI-era company forms into 6 types, from small to large, from human-driven to AI-driven:

A: Solo + AI Fleet (1 person)

One person managing a fleet of AI agents covering product development, growth, support, and ops.

This is where I am now. Revenue ceiling roughly $100K-$5M — depends on the product. Best suited for MCP Servers, CLI tools, APIs, newsletters, and autonomous agents — products with low build cost (see Level 1-3 product forms in the Stack Ownership Spectrum).

Where the ceiling is: Enterprise clients don’t trust it, 24/7 response is impossible solo, compliance is a blind spot. Pieter Levels (Nomad List, RemoteOK) and Marc Lou prove the ceiling can be high, but they build B2C or developer tools — no enterprise sales needed.

B: Micro Elite Team (2-5 people + AI army)

Each person owns a domain, each managing 10-50 AI agents. My thesis: this may be the most efficient configuration for the next decade — because it’s the smallest unit that can cover enterprise sales.

Revenue ceiling $5M-$50M. What does it add over solo? Enterprise sales capability (at minimum CEO + sales), 24/7 coverage (across time zones), compliance handling. One person building product + AI executing is already powerful; adding 1-2 complementary people (sales, design, domain expert) dramatically expands addressable market.

HBR is already discussing the “Agent Manager” role — someone who manages AI agent teams like a PM manages products. This role is especially critical in 2-5 person companies: everyone is both decision-maker and agent manager.

C and D: Scaling Paths (10-1000 people)

Further up, it’s traditional companies with AI upgrades. Type C (10-50 people) replaces today’s 500-person company — humans do strategy and client relationships, AI handles all execution. Revenue ceiling $50M-$500M. Type D (100+ people) is the platform giants’ battlefield — Anthropic, OpenAI, Google. Indie developers don’t play here, but your product might distribute on their platforms.

For most readers of this article, A and B are the real decision space. C and D are reference points for “if your product must scale to that size.”

Looking Ahead: Zero-Person Companies and Agent Networks

More extreme configurations already have prototypes. Autonomous companies (0-1 person) — humans are just shareholders, AI runs everything — simple versions exist today: algorithmic trading, automated dropshipping, programmatic SEO farms. Complex versions (autonomous SaaS) probably aren’t viable until 2030, with unresolved legal questions (can AI sign contracts? who’s liable?).

An even more radical idea: Agent Networks — no “company” entity, multiple AI agents collaborating via protocols with automatic revenue sharing. This needs agent-to-agent protocols + automatic payments + reputation systems, timeline 2030+. But the concept is worth watching — it means “company” itself might be redefined.

The Most Dangerous Position: The Middle Layer

This was my most surprising finding: the most dangerous position isn’t any single form — it’s the “middle layer” of 20-200 person companies.

Traditionally, mid-size companies (20-200 people) are the backbone of employment and GDP — not as agile as small teams, not big enough for platform effects, surviving on “big enough to take on work, small enough to control costs.” But AI is reshaping company size distribution into a barbell: many micro-companies + few giant platforms, with the middle hollowed out.

Why?

A 20-person company can’t outrun a 5-person elite team. The 5-person team crushes on decision speed, AI adoption speed, and cost of experimentation. A 5-person AI-first team can achieve $3M revenue per human, while a 50-person traditional team might still be struggling at $300K.

A 200-person company can’t beat platform giants. You don’t have Anthropic’s models, Google’s data, or Microsoft’s distribution. What you can do will either be built into big platforms or replicated by 5-person teams at low cost.

The historical reason for the middle layer was “coordination cost” — doing big things required many people. AI has dramatically compressed that cost. When one engineer + AI approaches the output of 5-10 past engineers (Cursor, Copilot, and similar tools are approaching this ratio), a product that previously needed 50 engineers can be built by 5 + AI. The middle layer’s reason to exist is disappearing.

This isn’t prediction — it’s happening. Many $100M ARR AI companies have fewer than 100 employees. Conversely, traditional SaaS companies are laying off, downsizing, “AI-transforming” — many going from 200 to 50 to 20. This middle-layer squeeze is most visible in software and professional services, because their core work is almost entirely information processing — exactly where AI replacement efficiency is highest.

The middle layer’s disappearance has a byproduct: “companies” are becoming “projects.” When creation and shutdown costs approach zero — one person builds an MVP in a week, validates the market in a month, decides to continue or quit in three months — the Pieter Levels model of running multiple products simultaneously becomes increasingly common. This aligns with the previous article’s logic: when build cost approaches zero, focus not on “what to build” but “at which layer do I have something others can’t take away.”

Apply the same logic to company form: don’t ask “how big should my company be” — ask “at the scale I’ve chosen, what irreplaceable configuration do I have?”

How to Choose? Start with 4 Questions

Don’t start with “how big do I want to be.” Start with these 4 questions:

1. Who is your customer?

• Individual consumers / developers → A is enough
• SMBs → B (need at least one sales role)
• Enterprise → Start at C (compliance + security + SLA are hard requirements)

2. What does your product need?

• Pure digital delivery (code, content, data) → A or B
• Physical touchpoints (hardware, offline services) → C or above
• 24/7 ops → Start at B

3. What irreplaceable resources do you have?

• Proprietary data → Any form works, data is the moat
• Proprietary technology → A or B, stay lean
• Brand/trust/relationships → Depends on industry, may need B or C
• Nothing → Start with A and accumulate

4. What’s your timeline?

• Need revenue now → A (solo + AI), start immediately
• 1-2 years to build a moat → B (find 1-2 complementary people)
• 5 years to scale → C or D (but validate with A first)

A Reference Matrix

Which products fit which company forms? This matrix comes from my product form landscape analysis:

Product Form	A (1 person)	B (2-5)	C (10-50)	D (100+)
CLI / MCP Server	✅✅	✅	—	—
API / SDK	✅	✅	✅	✅
Web SaaS (data)	✅	✅	✅	—
Web SaaS (general)	⚠️	✅	✅	—
Newsletter	✅✅	✅	—	—
Autonomous Agent	✅✅	✅	✅	—
Desktop App	✅	✅	✅	—
Bot	✅	✅	—	—
Hardware	❌	⚠️	✅	✅
AI Platform	❌	❌	⚠️	✅

✅✅ = ideal, ✅ = viable, ⚠️ = conditional, ❌ = not recommended

The pattern is clear: Forms A and B best suit products on the right side of the stack ownership spectrum (Level 1-3: CLI, MCP, API). This isn’t coincidence — low stack ownership = low build cost = small teams can execute well.

My Choice

Back to myself.

I’m currently Form A — solo + AI fleet. Products focused on CLI tools + MCP Servers + blog content. This configuration matches my resources: no proprietary data, no proprietary algorithms, but consistent output capability and depth of thinking about AI product strategy.

Next step? If a product starts attracting enterprise demand, I’ll consider upgrading to Form B — finding someone good at sales to team up. Not to “go big,” but to cover what a solo company can’t do.

That’s my understanding of company forms: there’s no “best form,” only the form that matches your product, customer, and resources. But one position is genuinely dangerous — stuck in the middle, with neither the speed of a small team nor the scale of a big platform.

If you’re in that position, now is a good time to downsize. Not layoffs — evolution. Let AI replace coordination costs, multiply each person’s leverage by 10x, until your team is refined down to the truly irreplaceable few.

As for the original question — solo or team? The answer: validate solo first, and let the market tell you when to add people.

---

FAQ

Q: Can a solo company really hit $5M in annual revenue?

Yes, but it’s not easy, and it depends on product form. Pieter Levels’ Nomad List + RemoteOK + PhotoAI reportedly generate over $3M annually. Key conditions: global market, natural growth channels (SEO / social / word of mouth), near-zero marginal cost. For products requiring enterprise sales, solo caps around $500K — because one person can’t simultaneously build product and do enterprise sales.

Q: What signals indicate I should upgrade from Form A to B?

Three signals: ① You’re losing deals because clients need response speed or compliance you can’t deliver solo; ② 80% of your time goes to things you’re bad at (e.g., you’re technical but spending most time on sales); ③ You’ve validated product-market fit and the growth bottleneck isn’t product but execution capacity. Two or more of these means it’s time to find people.

Q: Can AI really let a 5-person team do the work of 50?

In pure digital product domains, it’s already happening. Prerequisites: ① Team members are senior-level — AI amplifies capability, it doesn’t replace experience; ② You must intentionally design AI workflows, not just “let ChatGPT write code” but build agent pipelines. For anything involving physical delivery, AI can’t help much yet.

---

This is the second article in the “AI Era Product Strategy” series. The first is Stack Ownership Spectrum — which layer of the tech stack your product should own. If you find this useful, I share this kind of thinking on Twitter.