AI broke traditional infra - with Kyle from Depot.dev @ AIE Europe
Depot cofounder Kyle Galbraith explains why AI-generated code is overwhelming infrastructure designed around human output, how owning the entire CI pipeline changes the performance ceiling, and what teams will need when five engineers can operate like 500.
Kyle Galbraith is the cofounder of Depot, which builds high-performance infrastructure for software development. He joined Scaling DevTools at AI Engineer Europe shortly after the launch of Depot CI.
The conversation moves from a sudden acceleration in customer workloads to a larger infrastructure reset: source control, CI and code review must all adapt when coding agents multiply the volume of changes moving through them.
The Q&A below is experimental — extracted from the episode transcript with AI. If you spot a mistake, let us know at jack@scalingdevtools.com.
What had Depot just launched?
Depot CI is an engine built from the ground up rather than another runner attached to somebody else’s orchestration layer. Its first supported syntax is GitHub Actions, while the longer-term design includes TypeScript and Python SDKs for expressing pipelines directly. The system runs on Depot’s own sandbox technology, giving the team control over the infrastructure beneath the workflow as well as the workflow itself.
What is changing in the way teams write and deploy code?
Depot’s growth rate had more than doubled in three months, which Kyle attributed to agents generating far more code and therefore far more builds. A developer may now produce ten times the former volume, but the surrounding source-control, testing, review and deployment systems were sized for human output. The bottleneck has moved out of the editor and into every system downstream of it.
Why is traditional developer infrastructure crumbling?
Most of it evolved over roughly 20 years with a human expected at each step. Even if a model can produce code almost instantly, that advantage disappears when a build queues, a test suite crawls or review cannot absorb the change rate. Kyle’s extreme thought experiment is a ten-person team directing 10,000 agents: infrastructure built around ten human contributors cannot survive the resulting load.
What has to change for agent-scale development to work?
The whole development loop needs to be faster and more programmable, not just the code-generation step. Teams need isolated environments, source control, CI and review systems that can coordinate large numbers of concurrent tasks without making engineers wait or manually shepherd every transition. The goal is infrastructure that lets humans supervise outcomes while agents handle far more of the execution.
Why build a complete CI engine instead of faster runners?
With accelerated GitHub Actions runners, Depot controlled only about 30% of the pipeline; the rest still depended on GitHub’s orchestration and its constraints. Depot CI lets the company own the full path and optimize it as one system. That matters when incremental runner improvements are no longer enough for the order-of-magnitude increase in work that agents create.
Where does Depot go next?
The team wants Depot CI to accept more languages and make its syntax layer pluggable, including community support for systems such as CircleCI and Jenkins. Kyle also sees an opportunity to extend Depot-level performance into source control and code review. The broader question is what collaboration should look like over the next five to 20 years when a five-person engineering team can produce like one with 500 people.
How does Kyle see the European technology scene?
London remains Europe’s clearest technology hub, while Berlin, Barcelona and Amsterdam are developing meaningful ecosystems of their own. Europe is not simply a smaller San Francisco: it has distinct research depth and local strengths that can produce important companies. The opportunity is to build from those advantages rather than judge every city by a single Silicon Valley template.