The Bug Backlog That Never Gets Fixed
Every engineering team has one. The small bugs. The annoying ones that are real and consistently affect users but are never urgent enough to displace whatever is on the roadmap. They accumulate. The developers see them. Nothing happens.
Yhangry is a private chef marketplace processing about £15 million GMV. The founder was on maternity leave when she decided to solve the bug backlog problem. She built an autonomous bug fixer in less than four days.
Week one: 25-plus bugs fixed and shipped.
One-shot fix rate: 60 to 70 percent on first attempt.
Why 60 Percent Is the Right Number to Care About
The benchmark of 60 to 70 percent one-shot success on real bugs is worth thinking about carefully. It sounds like a failure rate , 30 to 40 percent of attempts do not work on the first try.
But compare it to the alternative: zero percent, because no developer was ever going to get to those bugs. The ones that are not fixed on the first try still exist in the queue for a second attempt. The ones that are fixed on the first try shipped without human involvement.
The measure is not whether AI is better than a senior developer working carefully on a well-defined problem. It is whether it is better than nothing, applied to the category of work that was going to stay as nothing indefinitely.
The Patterns That Spread Across the Company
The bug fixer was the first agent. It did not stay the only one.
Every area of the company now uses agents or is restructuring to be AI-native. The founder described the spread as fast and somewhat organic , once one team sees what an agent can do with a specific workflow, they want the same for theirs.
One particularly specific use case: the founder takes screenshots of competitor product slides at conferences and sends them to Claude Code to analyze later. Everyone else at the conference is manually taking notes or following up with decks. She is building a searchable archive of competitive intelligence that the rest of her team can query.
The Product Vision This Led To
The internal agent work changed how the company thinks about its core product. Yhangry's existing flow , customer wants a private chef, searches, finds some, back-and-forth over several days, eventually books , started to feel structurally wrong.
The insight from running agents internally: chefs send the same information to different customers over and over. Customers ask the same questions of different chefs. All the matching data exists. The back-and-forth takes days because no one has connected the data to a system that can make the match instantly.
"It's like Claude for chefs." That is the product they are building: an AI that handles the entire matching and admin layer, so chefs can focus on cooking and customers can focus on the event rather than the logistics.
The founder got hooked on OpenClaw on maternity leave. Three months in, she was nearly separated from her phone. The company has never looked the same since.