Carmentara · Service line
Copilot, Cursor, Codex, Claude Code — seats approved, budget signed, pilots launched. Nine months in, cycle time is flat, the board is asking, and your best engineers are quietly reviewing AI output they didn’t ask for. We find the real bottleneck and make it move.
Engineering orgs, 50–500 developers · 2–3 engagements per quarter
Core thesis
The model is not your bottleneck. Your specs, your review loops, your ownership structure, and the half of your repos nobody has read in a year — those are. Until the upstream changes, every new tool just makes more noise faster. That’s the work.
Field notes
Three patterns show up in almost every engagement. We won’t tell you whether yours has them until we look.
Pattern 01
There’s a layer every team ships into production before their code ever reaches the model. We audit it; they don’t. The same misconfiguration shows up in that layer roughly 7 times out of 7. Fixing it takes an afternoon. Nobody was looking at it.
Pattern 02
The things leaders use to judge a repo’s readiness — test coverage, recency, documentation — don’t actually predict it. Something else does, and most orgs don’t track it. Clean-looking services fail. Legacy corners win. The gut read is backwards.
Pattern 03
AI-era reorgs fail at the same seam, every time. It’s a single role boundary that looks reasonable on the org chart and collapses in practice. The engineers who notice it first are the ones who quit. By month two it’s a retention problem.
Engagements
Four shapes of engagement. Scope and price are set after the diagnostic call — not from a menu.
01 · Diagnostic
You walk out knowing which of your repos will absorb AI and which are actively burning spend, where the delivery curve is being capped upstream, and the single move that unlocks the most velocity. Fixed fee. Written deliverable. Exec-ready.
02 · Workflow
A ticket-to-merged-PR pipeline your engineers trust, piloted on your stack with measured outcomes. You leave with a defensible cost-per-PR, a review process that actually catches what matters, and a rollout path from pilot squad to the rest of the org.
03 · Topology
If you’re about to reorg around AI, you do it once. You leave with a defensible team shape, the role changes that survive month two, a 90-day transition plan EMs can actually run, and the talking points for the people whose jobs just changed.
04 · Advisory
Ongoing advisory for the VPE or CTO running transformation while still shipping. Hiring that selects for the work that’s actually left. Performance review that holds up when agents write half the code. Coaching for leading through specs and outcomes instead of line-by-line review.
Positioning
Workflow first. Organization aware. Stack-agnostic. Built for production, not theater.
Recent engagement
Series C SaaS, ~120 engineers. +50% meaningful throughput — measured in merged work that shipped, not raw PR count.
Anonymized on request. More detail under NDA on the diagnostic call.
Next step
You describe the delivery curve, the AI investment, and the shape of your team. We come back with the two or three places we’d look first — and whether this is work we should do together.
