When Agents Outnumber Humans

We built SaaS on a simple belief — users are human — and it's breaking.

Imperva published the number last year. 51% of internet traffic is now bots, not humans. Most people filed it under "interesting" and moved on.

I think it's the opening line of a much bigger story.

Automated traffic (crawlers, agents, AI systems doing work) already outpaces human activity. IBM and Salesforce project 1 billion deployed AI agents by end of 2026. One research firm projects 2.2 billion by 2030 at 139% annual growth. 80x the current count in four years.

The number of active agents on the internet will exceed the number of active humans. When that happens, every assumption SaaS was built on breaks.

---

SaaS multiples have compressed from 15x to 8x forward revenue over two years. The market is repricing companies whose value is metered by the number of humans paying a monthly fee.

Hybrid models (subscription floor plus usage upside) are growing at 21% median versus 13% for pure subscription. Seat-based SaaS is being marked down while consumption-based businesses are being marked up.

Most product teams just haven't looked at these numbers yet.

---

Most software was built for one user: a person.

The UI is the visible part — buttons, dashboards, onboarding flows — but the whole business model underneath shares the same assumption. You charge per seat. You build for the human clicking through. Your roadmap is organized around a persona.

You give her a name. Sarah, the marketing manager. 28-35, uses HubSpot, frustrated by manual data entry. You build features for Sarah. You measure activation by whether Sarah logs in.

That made sense when Sarah was the user.

The shift most product teams haven't made: your persona should be an agent running in someone's Claude Code, not Sarah. An agent with your API credentials in its environment, running tasks on a schedule. It's prospecting, enriching lead data across three sources, updating a Google Sheet, doing a CRM lookup, firing outreach, all before Sarah opens her laptop.

The agent doesn't need your onboarding flow, doesn't get confused by your UI, and was never confused in the first place. Confusion is a human thing.

Build for Sarah and you get slow flows designed for a cursor. Build for the agent and you get fast, reliable, machine-readable APIs for something operating at a completely different speed.

---

A solo founder wakes up and the work is already done.

One agent checked for new leads matching their ICP and enriched each one into a row in Airtable (LinkedIn, Clearbit, CRM, all combined) — no human clicks required. Another drafted personalized outreach for the top ten and flagged competitor pricing changes.

The rhythm is a heartbeat. Every 30 minutes, agents check in: new leads, emails to follow up, data to sync. They run in parallel, ten things simultaneously, each one's output feeding the next. For the cost of API calls, a single founder runs what used to take a team.

---

A YC company called AgentMail builds email inboxes for agents, not for the humans managing them. Each agent gets a distinct identity, a separate trust model, its own inbox. Okta launched non-human identity management. Stripe powers agent-initiated purchases directly inside ChatGPT. Retailers are building "Agent Entrances": API endpoints with structured manifests where agents check inventory, evaluate options, and complete a transaction without a human ever touching the page.

These are products with paying customers, not experiments.

---

Cloudflare spent years building bot management tools specifically to block AI crawlers. Last week they shipped a /crawl endpoint: one API call, whole website, returned as clean markdown or JSON. Built for training pipelines, RAG, and content monitoring. One of the largest CDN providers on the internet — roughly 20% of all web traffic runs through them — just built agents a door.

When a company that size changes direction, it's not a philosophical position. They looked at their traffic data and followed it.

---

Your TAM used to be calculated in humans.

How many people need this? Eight billion, growing at 1% a year. A mostly fixed pie.

If your product serves something agents need (data, tooling, specialized capabilities via API) and agents are growing at 139% a year, you're selling to a population that could 80x in four years. The agent-addressable market for most software categories is already larger than the human one, and the gap widens every year.

Most SaaS companies have an API. Most treat it as a secondary surface, something developers use rather than the product itself. The API is where your product actually gets used now. By agents in orchestration layers your customers built, doing work at machine speed, on schedules, in parallel. Price it for that. Build your roadmap around the agent running in someone's Claude Code, not just the person at a desk.

Consumption pricing is how you capture that value. The agent doesn't need a seat — it needs calls, compute, data access. The companies pricing for that now are building revenue models that scale with a population growing at 139% a year. The ones still counting seats are pricing for a market that's already being left behind.