The 700-Agent Company, The Web Is Tapped Out, and The Marketplace Extinction Map

the agents are already here.

Good morning

In today’s edition, among other things:

• The 700-Agent Company

• The Case for Decision Traces

• The Web Is Tapped Out. Expert Data Is the New Oil.

• The Math Behind “Building in Public”

• The Marketplace Extinction Map

• Vertical AI Isn’t Competing for Your Software Budget

Onwards!

---

The 700-Agent Company

Notion has 1,000 employees and 700 AI agents. That ratio tells you more about the next five years of company building than most fundraising decks will.

Ivan Zhao’s recent post frames AI through industrial metaphors: steel, steam, bicycles, cars. The historical parallels are nice, but the concrete data point is what matters. A $10B company is running with roughly 0.7 agents per employee, and Zhao calls this “baby steps.”

Alongside our 1,000 employees, more than 700 agents now handle repetitive work. They take meeting notes and answer questions to synthesize tribal knowledge. They field IT requests and log customer feedback. They help new hires onboard with employee benefits.

The list is instructive: organizational sludge that scales linearly with headcount. Onboarding, IT tickets, meeting notes, status updates. Notion automated the communication overhead that makes companies slow, not their core product work.

AI is absorbing the coordination tax.

Zhao identifies two blockers for broader knowledge work automation: context fragmentation and verifiability. Code agents work because everything lives in one place (the IDE, the repo) and outputs can be tested. Knowledge work is scattered across Slack, docs, dashboards, and someone’s memory. You can verify if code compiles. You can’t easily verify if a strategy memo is good.

Today, humans are the glue, stitching all that together with copy-paste and switching between browser tabs. Until that context is consolidated, agents will stay stuck in narrow use-cases.

Notion cares about being the single context layer because if you’re building agents, the company that owns consolidated context wins. Fragmented SaaS stacks become a liability, not a feature.

Zhao’s critique of current AI adoption:

We’re still in the “swap out the waterwheel” phase. AI chatbots bolted onto existing tools. We haven’t reimagined what organizations look like when the old constraints dissolve.

The waterwheel metaphor comes from early steam adoption. Factory owners replaced water wheels with steam engines but kept everything else identical. Productivity gains were minimal. The breakthrough came when they redesigned entire factories around the new power source.

Most companies are doing the AI equivalent of swapping the waterwheel. Same processes, same meetings, same approval chains, plus a chatbot.

The operational implications:

1. Audit your coordination overhead. What percentage of employee time goes to syncing, updating, reporting, meeting? That’s your agent opportunity space.

2. Consolidate before you automate. Agents struggle with fragmented context. If your work lives in 15 different tools, you’ll get 15 mediocre point solutions.

3. Design for supervision, not intervention. Zhao references the 1865 Red Flag Act, which required someone to walk in front of every car waving a flag. Human-in-the-loop often means human-as-bottleneck. The goal is humans supervising from a leveraged point.

What would change my view: evidence that highly fragmented tool stacks with narrow AI solutions outperform consolidated platforms with deep integration. I haven’t seen it yet.

Zhao’s co-founder Simon went from “10× engineer” to managing multiple coding agents simultaneously, queuing tasks before lunch and letting them run. That’s the shift from individual contributor to orchestrator. The 700 agents at Notion are just an early signal of where headcount ratios are heading.

---

The Case for Decision Traces

Everyone agrees agents will change enterprise software. The debate is over who wins: incumbents that own the data, or startups that own the execution path.