Depot

The argument for building our own CI engine instead of wrapping GitHub Actions: When you wrap someone else's execution model, you inherit their constraints. Sequential steps. Queue times. Opaque resource allocation. We wanted parallel by default, instant cold starts, and real observability. You can't bolt those onto an architecture that wasn't designed for them. So we built the engine from scratch and kept the workflow syntax everyone already knows. https://lnkd.in/dxw6b5ZS

You spend more time waiting for CI than you spent writing the code it's testing.

Push. Wait. Guess. That's the CI experience for most teams: You push code. You wait 8-15 minutes. If it fails, you guess what went wrong from 400 lines of logs. You push a fix. You wait again. We think CI should work more like your local dev environment -- fast feedback, real debugging tools, and no queuing. That's what we built Depot CI to be. Jobs start in seconds. Steps run in parallel. You can SSH into a running job and poke around. Push, see results, keep moving.

Your CI steps don't depend on each other. Why are they still waiting in line?

agents write code in seconds. CI gives feedback in minutes. that ratio is the entire problem.

If Depot CI was an apartment listing. 🔥

Every CI job that starts with "install dependencies" is wasting time doing the same work over and over. Depot CI custom images let you snapshot your entire sandbox environment — tools, packages, dependencies — and reuse it as the starting point for every job. How it works: 1. Run your setup steps on the Depot base image 2. 'depot/snapshot-action' captures the full sandbox state and pushes it to the Depot registry 3. Reference the image in any workflow — jobs start from your snapshot instead of a blank slate No Dockerfiles. No separate build pipelines. Rebuild the image only when dependencies change. Docs: https://lnkd.in/ds_vgMKg

Depot CI tells you why your job failed. It reads the logs, shows the failing step, summarizes what went wrong, and suggests a fix. No more scrolling through 400 lines to find the one that matters. https://lnkd.in/esgSQPbE

AI agents don't care about your CI dashboard. They care about three things: 1. Can I trigger a build programmatically? 2. How fast does it finish? 3. Can I get structured results back? Most CI systems were designed for humans staring at a UI. That worked when humans were the only ones pushing code. Now agents push code 10-50x more frequently than humans. They need an API, not a dashboard. They need results in seconds, not minutes. Depot CI was designed for this from day one: API-first, per-second billing, parallel by default. This isn't a feature. It's an architecture decision.

The five stages of CI grief.