Halcyon builds developer tools for teams shipping code faster with AI assistance. When they decided to add an in-product coding agent, engineering estimated six months to build the infrastructure it would need to run safely in production.

The Challenge

Halcyon was three sprints into building agent infrastructure — model routing, retries, observability — none of which was their actual product. Every week on plumbing was a week not shipping the feature customers were actually asking for.

What they'd already built

By the time they evaluated Sublayr, Halcyon's team had a working prototype of routing logic and a basic retry wrapper, but no observability layer, no approval flow for risky actions like committing code changes, and no cost tracking at all. Estimating the remaining work honestly put them at another four to five months before anything resembling a production-ready system existed.

Why the timeline mattered

Halcyon was competing directly against two other dev-tools companies racing to ship agentic coding features. A six-month infrastructure build wasn't just an engineering cost — it was a competitive risk, handing rivals a multi-quarter head start on a feature category everyone in the space knew was coming.

The Solution

They moved their coding agent onto Sublayr: any-model deployment, approvals on merges, and full run traces out of the box. No infrastructure to build.

Migrating the existing prototype

Halcyon's existing agent logic didn't need to be rewritten — only the infrastructure layer around it changed. Their prototype's core reasoning stayed intact; Sublayr replaced the custom routing and retry code they'd built with a managed equivalent, and added observability and approvals neither of which had existed before.

Approvals on code changes

Because the agent could open pull requests and merge changes autonomously, Halcyon configured an approval gate on any merge to a protected branch, routed to whichever engineer owned that repository. This let the agent handle routine changes end to end while keeping a human in the loop on anything touching production code paths.

The Results

• 6 months of infra roadmap cut

• 14 days to agents live in production

• 60% less infra code to maintain

• 99.99% uptime from day one

What changed for the engineering team

Beyond the headline numbers, the team reported a shift in how they spent their time. Engineers who had been assigned to infrastructure work moved back onto product features within the first month, and the team stopped fielding ad hoc "why did the agent do that" questions internally, since every run was now traceable by anyone on the team, not just whoever happened to build the original prototype.

In their words

"Sublayr is the infrastructure we were about to spend half a year building." — Tom Alvarez, CTO, Halcyon