Rescuing a stalled build

Rescue work is a specific skill, and it isn’t the one people assume. When a build stalls — the release that keeps slipping, the codebase nobody wants to touch, the project that’s burned through two teams — the fix rarely starts with more technical firepower. It starts with recognition.

Here’s the counter-intuitive part: the most valuable qualification for rescue work is experience failing. Decades of trying things — sometimes succeeding, sometimes watching an approach collapse in a way you didn’t see coming — build a catalog that no amount of raw talent substitutes for. You’ve seen how projects actually die: the abstraction that metastasized, the dependency nobody owns, the “temporary” workaround holding up half the system, the team that’s been building around a problem so long they can’t see it anymore.

That catalog is the skill. Walking into a stalled project, the question isn’t “am I smart enough to fix this?” — it’s “where do I start looking?” Breadth and depth of experience mean the common failure modes announce themselves, and the uncommon ones at least suggest a direction. That’s a skillset a lot of very good developers and teams simply don’t have yet, through no fault of their own. You can’t read your way into it. You have to have shipped — and stumbled — enough times.

What’s changed recently is what happens after the diagnosis. Rescues used to force a triage mindset: fix what’s fatal, patch what’s ugly, leave a list of “shoulds” nobody will ever fund. If you know how to wield the leverage AI actually provides — the production kind, not the demo kind — that trade-off softens. The same recovery now delivers results faster — and goes further: better quality, more thorough hardening, the improvements that used to be left on the list actually getting done. Experience decides what to fix and in what order; the machine makes it economical to fix it properly.

Stalled doesn’t mean doomed. A typical rescue here runs about 90 days. What it usually needs is someone who has seen this exact failure before — and, these days, has the leverage to do something about all of it, not just the fatal parts.

Got one that won’t compile, literally or otherwise? That’s the work we like best.