FairBakeWhat's slowing your engineering team down?
I've seen reviews pile up behind coding agents' output, quality slip as more code goes unread, and a few engineers pull far ahead while the rest fall behind.
Tell me where your team is stuck. In 30 minutes I'll tell you what's worked at other teams and what I'd try first.
Ari Wilson18 years building software, at Google, Celonis, and VideoAmp
You leave with
- 1
Where your team is likely losing time with agents, in the tools and the process around them.
- 2
What other teams have done about the same problem, and one thing to try first.
- 3
A follow-up with advice tailored to your main problems.
Who you'd work with
I'm Ari Wilson. I've spent 18 years building software and the engineering orgs that ship it, across advertising, big data and B2B. At Google I spent 13.5 years in ads. I led the teams behind Search ads forecasting, a system making millions of ML predictions a second over petabytes of data, then managed about 30 engineers building YouTube and App ad planning. At Celonis I started Los Angeles and US engineering, led 40 people across five teams in the US and Europe on enterprise process-intelligence products, and started a developer tools team. At VideoAmp, as VP of Engineering, I oversaw AI infrastructure and all of the company's user-facing teams.
What I've measured
Bakeoff is my eval. It reruns bug fixes a team already shipped through every coding agent they're weighing, graded against the tests that proved each fix.
An agent's default can be its worst setting
Fable 5.1 ships at high effort in Claude Code, where it solved 75% of tasks at $3.56 a run. At low it solved 88% for $2.33. Only 2 of 5 models solved more when turned up.
Higher effort doubles what reviewers read
On runs that worked, the mean diff grew from 126 lines at low effort to 281 at high. At 10 agent PRs a week, that's about 5 more hours of review.
Start with a call
Tell me where coding agents are slowing your team down, in the tools or in the process around them. There's nothing to prepare.