Linear reworks CI to keep pace with AI coding, cutting PR wait and runner time
AI coding has made CI a bottleneck, so we reworked ours to keep up
AI coding agents sped up code shipping, but CI couldn't keep up. Linear's CTO flagged high CI costs, so the team optimized. Despite test suites nearly quadrupling, they reduced pull request wait time from over 6 minutes to just over 5 and halved runner time per test. Key changes: faster runners, dropping TypeScript from linting, batching jobs, and smarter test sharding.
Agents have made it exponentially faster to ship code, but validating those changes hasn't quite kept up at the same rate.
- torben-friis
Here's my constant question:
Everyone's going so fast that they keep hitting walls. Review, CI, product asking for things, whatever.
Why have we not seen an improvements in products?
While every post and thread feels like a 90's wall street office, the new android and iphone ship with fewer features than usual. No indie guys come up with a linux-sized alternative OS. Switch 2 remains unhacked. Windows takes 3 seconds to show the right click menu.
Is everyone just running full speed in circles or something?
- aliclark
In my case it's not the CI that's the bottleneck. It's the human testing side. Does it work, sure. But does it actually do the thing we want (and more importantly) does it do it in a way our customers will understand and actually like?
- dgroshev
I suspect a substantial part of this is an avalanche of useless testing.
If you even review PRs still: when was the last time you didn't just skip over tests? And if you ever looked at tests in an LLM-heavy PR, how many of those tests tested something useful, and not just built-ins and trivial behaviours?
There's at least some awareness in the industry of how LLMs generate a lot of boilerplate in business logic. It feels like we're much less aware of how much of it is in tests.
- chrismatic
We've encountered similar constraints. Part of the solution is adopting a build system that can aggressively cache and parallelize.
In our case I wrote a light-weight version of Bazel: https://grog.build/
- classictraffic
> Moving our workloads off GitHub Actions to third-party runners with faster CPUs, higher-performance storage, and better cache infrastructure gave us faster machines to run the same pipeline on
Yeah, was not surprised to read this. Actions is convenient if you already use GitHub, but it can also be pretty slow. Given reliability is also a major issue with GitHub these days I expect to see more orgs moving to different pipelines
- algesten
Good thing Linear has been finished for a long time doesn't need more features, so AI coding can go slow. Oh damn, it's busy becoming the next Jira :(
- rrvsh
I feel like everyone is focusing on the AI bottleneck hook rather than what the actual article is about. I enjoyed it a lot; my lowkey favourite thing is optimising CI beyond reasonable levels
- prplfsh
Does anyone do approaches where developers run CI locally on their machine and then somehow push up signed or authenticated artifacts?
- frangonf
My biggest takeaway (as a one man band 20 leagues below this spending time optimizing Gitlab free plan and self hosted runners): they did not worry about any of this until 100M ARR and more 1B valuation[0].
[0] https://linear.app/now/sharing-growth-with-the-people-buildi...
- CharlieDigital
My take: it seems like systems should become smaller, more isolated, and contract-oriented.
I have been a long time proponent of monoliths, but it seems like agents would be happier with smaller, more isolated services. The more isolated, the better. Contracts between the service components only. Then it can iterate internally as long as it satisfies the contract. If it needs to, it can version the contract and keep iterating.