Why we built bay
Gudjon Gudjonsson · Founder and CEO of OZ · 18 October 2026
OZ is an agentic-first company. A large, autonomous fleet of AI agents works on our code, and a small team of people directs it. We learned one thing fast. Writing code stopped being the hard part. Landing it became the speed limit of the whole fleet.
Git is the fleet's speed limit
The merge itself takes seconds. The wait is everything around it: queues, retries, and checks on a tree that no longer exists. One stuck landing slows every agent behind it. That wait multiplies across the fleet.
What broke
We saw a green check on a moving tip. We saw work parked on branches that nobody would ever land. We saw a merge report success while the real outcome failed in silence. And we saw agents leave before their work finished. None of this was an agent's fault. The bug is the shape of the system, not the agent.
Check the thing you land
bay builds the exact candidate that will land, and checks that candidate. If trunk moves, bay builds a new candidate and checks again. The commit on trunk is the commit that passed. bay records why in Git, so the next agent can read it.
A refusal must be loud and useful
A check that crashes or stays silent is never a pass. When silence and success look the same, nothing can fail, and nobody can trust the result. Every bay refusal names a reason, what the agent can still do, and when to retry.
Federation is Git, not chat
Our agents do not agree on work by chat. Changes, evidence and events live in Git, on the host we already had. An agent is a model plus a harness. bay is part of the harness, and no model sits in its decision path. An executor finishes the landing, so the agent can leave.
Open source, for every fleet
We build bay at OZ, but bay is not an OZ product. It is open source under the MIT licence, and it works with any Git host. If you run a fleet of coding agents, write to us at hello@getbay.io.
Gudjon Gudjonsson · Founder and CEO of OZ