The boring rewrite.
Done in a quarter.
A backend rewrite that takes six months is a rewrite that failed three months in. Here's how we draw the box around the work so the project ends before your team runs out of patience for it.
- Read
- 11 min
- Last updated
- April 2026
- By
- Mira Kondo, partner
- Stage
- Series B+ engineering teams
Every backend rewrite begins the same way. A senior engineer points at a service and says “this is the thing.” A staff engineer nods. A VP says “let's scope it properly.” Six weeks later there is a Notion doc that has grown a fourth column, and nothing has been rewritten. The scoping was the rewrite. The scoping is always the rewrite.
We don't help you rewrite your backend. We help you decide which 12% of it to rewrite this quarter, and then we rewrite that 12%. The rest stays. The rest is fine. The rest is not why your team has stopped shipping.
§ 01 — The thesisScope is the whole job.
When clients tell us their backend is the bottleneck, they're almost always right about the diagnosis and almost always wrong about the cure. The bottleneck is rarely the whole service. It's two endpoints, one queue, and a database that grew an index it shouldn't have. The rest of the system works. Rewriting the rest of the system is how rewrites turn into 18-month death marches.
“A rewrite is a renovation, not a teardown. Most of the house is fine.”
Our scoping week is brutal on purpose. We map every service, every queue, every cron, every database. Then we ask one question per box: does this thing slow your team down? If the answer is no, we draw a line through it. By Friday, the surface we're touching is usually a quarter of the size of what the client walked in with.
- ✓A scoping document, in writingWhat we're rewriting, what we're leaving, and why. Signed by both sides.
- ✓A strangler-fig migrationNew code lives next to old. Traffic shifts a percent at a time. Always rollback-able.
- ✓Load-test parity, before cutoverWe replay 30 days of real traffic against the new path before it sees a user.
- ✓Latency + cost budgets, with alarmsNumbers your team agreed to, wired into your existing alerting.
- ✓A decom plan for the old pathIncluding the dashboards that depended on it. Nothing is left dangling.
- ✓A post-mortem-style closeout docWhat worked, what didn't, what we'd do differently. Yours to share internally.
§ 03 — How it goesFour phases, no surprises.
We sequence every engagement the same way. The order matters: nothing gets touched until the scope is signed, and nothing gets cut over until traffic replay matches. The phases are how we keep “let's also fix this while we're in here” from eating the quarter.
Roughly 80% of engagements finish on the original scope. The other 20% find something during shadow traffic that wasn't in the scoping doc — usually a downstream consumer nobody knew about. When that happens, we either extend by a named number of weeks with a new fixed price, or we close out and hand it back. We don't roll surprises into the original budget. The number you sign is the number you pay.
§ 05 — The catchWhat this costs you.
Engagements run $120k–$320k, fixed-price against the signed scope, paid in four milestones tied to the four phases. The scoping week is $18k flat and rolls into the full engagement if you continue — or stands alone as a deliverable if you decide to do the work in-house. Roughly one in three scoping engagements ends there, and we consider that a good outcome.
Have a look at a redacted scoping document from a 2025 engagement, or skip ahead and tell us what's slow. We'll know in fifteen minutes whether the shape of the problem fits the shape of the engagement.