T3 Stack creator Theo Browne argues Git is no longer the correct primitive for development and calls for an overhaul

4. Source control shouldn't require a "real os" + file system.

In a world with tools like "just-bash", it's silly that we need a full kernel and file system to even interface with git. Reading and updating files should be doable with simple API calls.

3. Worktrees are an abomination

We need source control that makes more assumptions about our file systems. Why do we have to implement copy on write ourselves? Why can't I "check out" the same branch in 2 places at the same time? Why do I have to put so much energy in keeping my worktrees updated to main?

We've been taken for fools. The current worktree implementation in git should be ignored entirely. Our agents deserve better.

1. "open source" should not always mean "100% of our code is public 100% of the time"

How much energy have we put into preventing .env leaks in our source control? How many miserable ways have we re-invented env var sharing?

How many projects would be open source if they could hide in-flight PRs?

How many security fixes are sitting unpublished because they will be exploited as soon as they appear in the tracker?

How much better would life be if I could have a monorepo with some sub-packages that are "private" without splitting into multiple repos?

2. Commits are a bad primitive. Branches are worse.

jj gets much closer to "right" here. Work is constantly staged as it is done. Snapshots are taken when you run any command at all. "commits" force us to think about our history at all times. That's stupid and a waste of our time.

We should be able to just edit the code. We can worry about how it is "tracked" later.

@theo All 3 points are correct and solid, yet I fear the institutional momentum of git is so strong it can never be replaced, akin to email.

I have many more thoughts but I want to get back to work.

If you take anything from this, please let it be this:

git isn't perfect. It's barely even good. We've dealt with it for decades so it's hard to see the cracks. Be more introspective. We can do so much better.

@theo most git friction isn't the tool, it's merge workflows creating sequential bottlenecks. the real question is what primitive replaces branching without making collaboration chaos. what are you thinking?

I know exactly how it sounds to reply to "someone should build this" with "mate I've been building this", but git is not the right primitive is the thing I've been screaming into a text editor for months, so here goes. The writeup is the argument in full (files flatten a graph of intent into text, two agents land clean merges that disagree at runtime, retries aren't idempotent because git was built for one human at a terminal): the next software stack needs more than code generation, https://dev.to/wernerk_au/the-next-software-stack-needs-more-than-code-generation-3aep

@theo clouds for agents = ?

didn't cloudflare already launch something like this?

@theo Branches are not a bad primitive as long as the agent runs in an sandbox env which is tied to the branch

Launching the same sandbox for a PR means that chats/commands and all history are stored in the sandbox/PR.

So they are still auditable but die when it's no longer relevant.

@theo Every Claude Code session I run ends with 30-40 micro-commits. The history is immediately noise. Git was built for humans reviewing patches, not agents that explore, backtrack, and retry. The audit trail primitive needs to be semantic, not line-by-line.

@theo Thus us what @tekbog did with clanker cloud

@theo is your cloud including the physical datacenter(s) + power + PoP rollout, geo co-locating servers, or some other deployment?

@tekbog @thoughtlesslabs This is the naive solution. Almost built this myself.

I've concluded that it's wrong, and the primitives exposed by existing clouds are not worth wrangling.

@thoughtlesslabs @theo haha yea just try http://clankercloud.ai

@theo @thoughtlesslabs why is it wrong? you give engineers an ability to do what they want with ai when they already have accounts, their companies have credits, etc

@theo are LLMs dying? will any of this matter this time next year?

@theo @thoughtlesslabs terraform is static, this is not, that’s the whole point, if you are still thinking about terraform then you need a better world model - code is liquid and infra should be the same

everything is bro2bro now aka agent2agent aka clanker2clanker

@theo You’ll get hate for this take but it’s extremely correct. git is, at best, “better than what was before”. It is not the best we can do.

@tekbog @thoughtlesslabs Because you are patching existing solutions that you assume are valuable enough to still optimize for.

There’s money here for sure. I just don’t care to reinvent terraform for the 50th time. I want to boil the fucking ocean.