Dan Manges

Building the next generation of CI/CD: @rwx_cloud. Previously built @RootInsurance (IPO) and @braintree ($800M acq)

[Philippe Noël @philippemnoel]

Conference alert: The incredible @isaacvando from our team is organizing Software Should Work, an independent conference about software reliability, on July 16-17. ParadeDB is not affiliated, although we are proud sponsors.

Speakers include the creators of Zig, SQLite, etc. More info: https://t.co/Ny73ZiyYsv

[Dan Manges @dan_manges]

@JustJake @solomonstre We built a replacement for BuildKit. Completely compatible with the OCI spec but radically faster. Unlike BuildKit’s lineage based caching, we implemented content-based caching. Not open source yet, but I’ll show you if you’re interested. Builds on Railway would be so much faster

[Dan Manges @dan_manges]

Is there more to this race than having hot pools?

[Malte Ubl @cramforce]

Vercel Sandboxes are now the fastest sandbox using real VMs as security boundary based on the @computesdk benchmark. The team has been absolutely cooking on this.

And the best thing: Because we have a unified Fluid Compute stack across Sandbox, Builds, and Functions these wins are often shared across the stack.

On the feature side there is a really exciting roadmap ahead as well.

My favorites (all driven by feature requests from our customers):
- Persistent sandboxes (in beta, GA immanent)
- The fully mutable firewall also becomes fully programmable

[Isaac Van Doren @isaacvando]

I'm enormously excited that Richard Hipp, the creator and lead maintainer of SQLite, will be speaking at Software Should Work. Join us on July 16-17!

[Dan Manges @dan_manges]

@FurqanR Try @rwx_cloud

rwx sandbox exec —

rwx run .rwx/ci.yml

Automatic filesystem syncing.

Close the loop for agents so they’re not pushing broken code.

Automatic content-based caching so you don’t have to deal with snapshots.

[Dan Manges @dan_manges]

@ryancarson @bernhardsson You should check out RWX. Run from a local CLI, results output from the CLI, automatic content-based caching, graph-based task model (deduplicates steps). Runs on normal container images but powered by a new container runtime. It’s the best CI/CD for agents.

[Isaac Van Doren @isaacvando]

Thrilled to have Andrew Kelley, creator of @ziglang speaking at Software Should Work. Early bird ticket sales end at midnight!

[Dan Manges @dan_manges]

@richardartoul We built this to do that. Which test framework are you using?

https://t.co/OYGlCoLZIe

[RWX @rwx_cloud]

"RWX was built by engineers for engineers, and it shows in how fast and hands-off CI feels day to day."

Learn how Tropic migrated from a bloated GitHub Actions workflow to RWX, cutting CI runtimes by 3x while accelerating the path from PR to deployment.

https://t.co/ynBsZDwfqw

[Dan Manges @dan_manges]

@lukerramsden I’ve been building a new CI/CD platform for a while now. It’s CLI-driven so it works super well with agents. It’s also the only platform with automatic content-based caching. I’d be happy to hop on a call if you want to see it.

[Dan Manges @dan_manges]

@alvinsng @FactoryAI if you really want to cook use RWX. agents can run without a git push and content-based caching means your agents will never be waiting longer than necessary

[Dan Manges @dan_manges]

Unreal speaker lineup at this conference. I’m going.

[Isaac Van Doren @isaacvando]

I'm extremely excited that Carson Gross, the creator of @htmx_org is coming to speak at Software Should Work. Just one week left for early bird tickets!

[Astasia Myers @AstasiaMyers]

Hot take 🔥: AI coding tools are about to make CI a big bottleneck in software development.

We saw it before:
AI → more code → code review became critical

Now:
AI → more commits → CI pipelines are collapsing

AI coding agents require a new CI system.

[Dan Manges @dan_manges]

@dboskovic Thanks for the tip! We’ll look into this.

[Dan Manges @dan_manges]

You can build your own CI (this is impressive) or you can use RWX

[David Boskovic @dboskovic]

coming right up but TLDR

Cut out all setup overhead (prev 30s, now about 1s)
- ovh amd epic turin box with 128 cores and 256gb ram (1k/mo)
- golden image of main with all cache loaded
- zfs for instant copy of golden image (this is magic)
- git fetch all every second for local mirror
- golden image of database so only last migration runs (as Postgres template)

- turbo cache locally

For actual suites
- much higher sharding since now no overhead to each shard
- use @bunjavascript tests where possible to avoid typescript compilation
- incremental typechecking with local cache

Bypassing GitHub actions in favor of custom check suites
- a few seconds of queue time saved
- no action minutes billed (we hit 36k minutes in 3 days)

For preview apps
- JiT full stack preview apps (not deployed on each commit)
- 2-3s cold start on any commit sha to a fully deployed full stack preview app
- zfs clone of golden firecracker vm and then check out latest commit etc

[Dan Manges @dan_manges]

@dboskovic We actually allow starting runs from the CLI without a git push. Content-based caching has the clone cached, the CLI sends up a patch, the rest of the cache pulls at 250 MB/sec. If you have 2 GB of dependencies you’re ready to go in 10 seconds, even with ephemeral compute

[Dan Manges @dan_manges]

@ankrgyl @aayush_shah15 I’ve been building a new CI platform from first principles. Automatic content based caching, graph based task execution, and agent-native interfaces (CLI and semantic outputs instead of git push and logs). I’d love to know what you think.

[Dan Manges @dan_manges]

“if someone else is always upstream, you can’t really redesign workflows or build anything that defensible”

[signüll @signulll]

while they require more capital fully verticalized startups like this are super fascinating to me cuz in the ai era the two things that matter are owning the underlying context & deploying intelligence on top of it.

most ai startups do the latter on someone else's platform because it's easier to sell, i.e. you don't have to ask customers to rip out the whole god damn stack. but if someone else is always upstream, you can't really redesign workflows or build anything that defensible.

in specific verticals where incumbent software is fragmented enough to allow rip & replace, you'll likely see more companies running the toast playbook which is to own the system of record then make every ai feature a retention mechanism rather than a standalone product.

[Dan Manges @dan_manges]

@jeff_weinstein we did. RWX is agent-native: run from the CLI instead of git push. get semantic, parsed errors in the CLI instead of downloading logs. automatic content-based caching to skip redundancy. when agents push the build is green and it passes instantly.