Light speed builds. Light weight bills. 10x faster Github runners, and 10x cheaper. Get started in 30s.
Joined January 2021
- Tweets617
- Following49
- Followers827
- Likes366
Pinned Tweet
My life's goal is to make this obsolete -
xkcd.com/303/
When we started on the Argonaut journey, we got 1-2 friendly customers while building out the MVP. But then, we stopped selling.
I was delaying launching with reasons like “we don’t have self-serve yet” or “this feature will make for a nice launch”. It took a while to come to terms with it, but it was fear of rejection for someone trying to sell something for the first time.
We are doing things very differently this time around, and learning from previous mistakes.
✅ Picked a problem that is narrow, and the value proposition can be conveyed in one sentence.
✅ Put up a website and a waitlist before development to gauge interest.
✅ Continuously selling - everywhere, all the time.
Everybody wants builds to be faster. Everyone wants costs to go down. The value couldn’t be clearer and we are seeing fantastic results.
Earlier today, a customer got setup within ~2 minutes and this is what they had to say.
I think we are onto something here 🔥
WarpBuild retweeted
Commits at agent speed lead to highly concurrent container builds at deploy time. Our remote docker builders now have completely revamped internals to support horizontal scaling while keeping the caches warm.
Once the primary builder hits capacity (1 build per 4 vCPU), a helper builder starts. On 16 vCPU primary, that’s the 5th build.
This is a transparent change rolling out to the entire fleet this week.
Commits at agent speed lead to highly concurrent container builds at deploy time. Our remote docker builders now have completely revamped internals to support horizontal scaling while keeping the caches warm.
Once the primary builder hits capacity (1 build per 4 vCPU), a helper builder starts. On 16 vCPU primary, that’s the 5th build.
This is a transparent change rolling out to the entire fleet this week.
WarpBuild retweeted
latest typecraft video is out!
youtu.be/URrSf2uKg0c
The shipping velocity at WarpBuild is through the roof! Here's what's new:
* Auto-scaling docker builders: docker builders automatically horizontally scale at higher concurrencies while retaining cache behavior
* `xfast` linux arm64 runners: 2-3x single-core performance on MacOS hosts.
* Runner metrics + logs to your own tools, via OTEL exports
* xcode 27 and iPhone Duo simulator
* Claude managed agents
* Security patches and image updates
* WarpBuild git checkout mirror for faster downloads at high concurrency
WarpBuild retweeted
1. Agents are going to eat the entire software development lifecycle
2. There is no upper bound to how much software we need
3. Tons of SaaS services are about to have their lunch eaten
In the future, software will be invisible.
WarpBuild retweeted
WarpBuild now has Linux ARM64 runners on Apple M4 Pro hosts.
PassMark single-thread:
AWS c9g (Graviton5): ~2,450
Apple M4 Pro: ~4,550
That is about 1.9x the CPU speed for your arm64 builds and tests.
Use the `-xfast` arm64 runners for your github actions and eliminate the wait.
warpbuild.com/docs/ci/cloud-…
WarpBuild retweeted
Everyone's posting about how CI bills are exploding.
That's true - here are some practical solutions for teams operating at scale.
- cache heavily.
- use the right sized instance. 2x instance size that reduces runtime by half still costs the same while reducing your wait time. obv not all workloads parallelize well.
- if CI reproducibility isn't important, consider using snapshots or starting from precomputed states for incremental builds. This is why your laptop build feels so much faster than a CI job.
- if building containers, use a heavily cached remote container builder. This is not the same as just uploading cache layers to a remote registry. Speed diff here can be ~10x between the two approaches.
- reduce number of tests. You know you don't care about some of them.
- use @WarpBuilds for faster single core performance and locally connected NVME drives which are the single most important factor for build and test workload performance.
WarpBuild retweeted
Shoutout to @WarpBuilds for making SSO easy. Fair price, easy enablement, no EnTeRpRiSe SaLeS cYcLe. Literally best in class!
Join us
@WarpBuilds is hiring engineers.
we're 8 people making ci fast for a lot of teams, with a new ai product in the works.
you'd solve big problems from day one.
please join us if you like solving hard problems on infra, scale, agents-human collaboration with a focus on developer and user experience.
warpbuild.com/jobs
Speed up your runners with WarpBuild.
It's faster AND cheaper, and takes one line to implement.
So I made a GitHub Actions workflow to compare 18 runner configurations to build RDM Windows from Windows or Linux, using GitHub runners and @WarpBuilds runners, and then graphed the results using Claude. Even with a future $0.002/min tax for self-hosted runners, it's worth it!👇
WarpBuild retweeted
天内迭代和分钟级发布，这个古法软件工程也可以做到。
ScopeDB 在新年左右大概三个月发了 50 个版本，这还不是早期快速迭代时的速度，而是内核相对稳定后维护和新功能逐步开发阶段。
Release Workflow 得益于 @WarpBuilds 的 SaaS 服务，我夸过很多次升级配置时间减半最终支出不变，只需十几分钟就可以发布。
到 CD 看不同环境，技术上也是一个 GitOps 触发的。
悲观者正确，乐观者前行。
宝玉老师提出了很多条质疑的点，本着严谨交流的态度，我不介意一一回复。
这篇文章内容涵盖了大量软件工程，因为我们希望把自己AI First如何实践落地的理念分享出来，而不仅仅是一些形而上concept的内容。至于其他AI First的实践，由于篇幅和文章重点无法涵盖所有。
第一，AI提交代码是不需要全部人工回归一遍的。我们会拆细每次PR的影响范围，基本都只会涉及某个功能模块，而不是大范围的修改，这在手写代码时也是基本的软件开发协作规范。
在符合软件工程协作规范下的情况下，AI提交的代码修改完全可以被自动化测试覆盖，不需要担心它会搞崩别的功能。
第二，我们中间的测试/审查/发布确实做到了全部自动化跑通了，所以我们做到了今天这个迭代效率。
第三，我们每次A/B测试的线上监控基础设施也都是完善的，有充足的数据支撑我们做判断。我建议你学习了解下statsig。
第四，你为什么会把大而模糊的任务丢给AI？这个问题非常奇怪，我相信如果你做过管理，也不会把大而模糊，自己都没想清楚的任务丢给员工吧。
第五，系统架构的设计是任何软件工程团队的基本功了，拿出这一条来悲观质疑真的很像在抬杠。
结论是，我们全都做到了。
关于你提到的claude code和codex团队是否有这么搞的问题，事实上我们就是观察到了claude团队极其夸张的迭代效率，以及OpenAI工程团队在今年二月份的分享得到的灵感：openai.com/index/harness-eng…
你觉得他们有没有也在用这一套呢？为什么你如此笃定AI交付和功能质量必然在对立面？
我们的分享来自于团队脚踏实地实践后的经验，关于我们目前做到了怎样的迭代效率，请看产品changelog：docs.creao.ai/community-and-…
WarpBuild retweeted
I've started testing @WarpBuilds custom GitHub Actions runners, and the first thing that stood out for me is that there's observability into runner resource consumption with recommendations. It's mind-boggling that GitHub doesn't have that, I've wanted this since forever
We're hiring. Links in QT
@WarpBuilds users using BYOC on GCP can now control the IOPS and throughput allocated to runner disks (we've supported this on AWS for a a while now).
It was fantastic to see this message from a user who noticed this and took the time to drop a message to us even before we could update the changelog.
We're hiring to accelerate our growth and build more such delightful capabilities for our users.
ycombinator.com/companies/wa…
ycombinator.com/companies/wa…
WarpBuild retweeted
⏰ BuildJet is shutting down on 31st March.
WarpBuild is the obvious choice to migrate to, because:
1. WarpBuild runners are faster for x86-64 and arm64 workloads.
2. WarpBuild has the broadest range of OSes and architectures supported (including Windows and MacOS)
3. WarpBuild has the most deployment modes , including Bring-Your-Own-Cloud on AWS, GCP, and Azure.
Switch now for faster and cheaper builds.
Here is the prompt you can provide to your favorite coding agent to make the switch:
```
Migrate runners from BuildJet to WarpBuild. The configuration for the WarpBuild runners can be found here: warpbuild.com/docs/ci/cloud-…
```
WarpBuild retweeted
We are the only GitHub actions runner infrastructure provider to support this feature: Snapshots.
Each CI job runs in its own ephemeral VM for isolation and reproducibility reasons. However, a significant portion of time in each job is spent on tasks that are repeated often. These are tasks like setting up dependencies, downloading artifacts that seldom change, etc.
WarpBuild's snapshots capture the full state of the VM at any point in the job's lifecycle and use that as the base for subsequent jobs. This has enormous potential because it can even cache things like partial compilations, test suites, etc. with zero effort by users.
One of our customers wrote a blog post about how they use WarpBuild snapshots in their CI to cut run times by 60-80% and save 24 minutes of CI time per commit.
Truly grateful to have fantastic customers like Andrew Berry. Read his article here: ddev.com/blog/ddev-ci-warpbu…
