For most of the last decade, agency capacity looked like a headcount problem. More clients, more deliverables, more people. The 2025 stack broke that math. A two-person team with the right tooling now ships more, on time, than most 8-person shops we compete with — not because we work harder, but because the loop is different.
The old math
An account manager fielding requests. A designer producing mockups. Two developers implementing. A QA person catching regressions. A project manager coordinating them. Five people, five inboxes, five sources of drift. Every handoff was a translation opportunity, and every translation lost information.
The new math
Two engineers, both able to hold the entire product in their heads. AI as leverage — for scaffolding, for research, for the boring parts of typed forms and CRUD screens. A CRM that reads incoming email replies with Claude and updates itself. Automation as a first-class team member, not a nice-to-have.
The result: fewer meetings (because there's no one to coordinate), faster iteration (because handoffs happen inside one person's head), and higher output per hour (because AI does the shape work). We ship more code than a team three times our size.
What breaks
Ambition. When you can ship faster, you're tempted to take on more — and the ceiling is real. Two people can't be in three time zones. Two people can't run a 24/7 support rota. Two people can't do enterprise sales cycles. The economics work if you pick engagements that respect the shape of the team.
The rule we run by
If a task can be templated, automate it. If it can't, ship it manually and watch for the pattern. Once the pattern is stable, automate that too.
That's how the sales-workflow CRM got built: after two months of manually classifying reply emails, we noticed the classification was 90% templatable. We wrote it up, gave it to Claude, and the loop closed.