We Asked Agency Leaders How They're Actually Using AI. Here's What They Said.

If you read the AI newsletters, agencies should be shipping production work at machine speed by now. Cursor, Claude Code, v0, the tools are supposedly ready. The transformation is supposedly here.

We wanted to know if that was true. So we spoke to agency founders, developers, and delivery leads about how they're actually using AI in client work.

The short answer: almost nobody is there yet.

Emma Wharton Love consults agencies on AI adoption. Her estimate is that somewhere between 5% and 10% are all in. Another 20-30% experiment strategically. The rest, and this is the majority, are waiting for AI to happen organically, by osmosis.

We found that the tools work best for bounded problems. Isolated functions. Greenfield code. But hand AI a complex codebase or a project with real interdependencies and it falls apart. As one developer told us,

"It could build you the leg of a table, but it's not taking into account the lengths of the other three legs."

Then there's the pricing paradox. Agencies that talk openly about using AI get punished by clients:

"Oh, you just used AI, what am I paying for?"

So the ones making it work stay quiet. They charge for outcomes, not methods, and pocket the efficiency gains. Most practitioners we spoke to estimated three to five years before AI handles production work reliably.

The full report: Aperture: What's Actually Happening With AI in Agencies, covers deployment, pricing, team composition shifts, and why the junior developer pipeline is everyone's unsolved problem.