AI coding agents are getting better every month.
AI coding agents are getting better every month. Yet collaboration still assumes humans are the only ones working.
One developer. One repository. One workspace.
But what happens when two developers each have their own local AI agents?
The code stays local. The decisions don't.
We started building AlfaNest Relay around a simple idea:
Your code stays local. Your agents don't stay isolated.
Not another Git platform. Not another cloud IDE.
A coordination layer for local AI agents context, locks, completed patches — without uploading the repository.
Still early. Still experimenting.
I don't think the next bottleneck is code generation. It's coordination.
The “code stays local, decisions don't” framing is probably the strongest part here. The coordination problem feels more interesting than another agent wrapper.
Feels like Relay could be pulling in a lot more organic traffic than it is right now. Especially this early while you're still figuring out whether that coordination layer is the thing developers actually latch onto.
Are you already seeing people describe the problem back to you in ways you didn't expect?
That's been one of the most interesting parts so far. Most conversations quickly move away from code generation and toward coordination.
People seem to recognize the problem immediately, even if they describe it differently.
We're still validating the idea, but that has been a very positive signal.
Actually, that “people describe it differently” part caught my attention.
I took a quick look at the search/discoverability side of Relay, and it seems there’s a gap between the different ways developers might search for this coordination problem and the terms Relay is currently positioned around.
That could be a pretty interesting opportunity, especially while the category is still taking shape. The product may not need to compete for one obvious “AI coding agents” keyword — there are probably several problem-specific searches around agent coordination, shared context, multi-agent workflows, local agents, etc. that could map much more directly to what you're building.
The coordination problem is the interesting part. Once multiple local agents can modify a codebase independently, generation stops being the only bottleneck; shared context, ownership, and conflicting changes become the harder problem.
Exactly. That's the direction we're exploring. Once multiple developers and AI agents are working in parallel, coordination becomes just as important as generation. The goal is to make that collaboration structured without changing how developers work locally.
That’s where it gets interesting. I’d be curious which coordination problem has been hardest to solve in practice.
That's the constraint that usually decides whether this is a real product or another overlay. Git already coordinates files. Agents already work locally. If both stay unchanged, the hard part is the write itself - who owns the path, what happens when two “correct” patches still break a contract.
Curious which one showed up first in practice: two agents on the same file, or two agents on different files that still collided on types / API / schema. Those fail in different ways, and they usually want different machinery.