I recently ran a pull request analysis for a company, looking at their last 1,000 pull requests. Ninety percent were fixes. Bugs, regressions, patches to things that had already shipped. Only one in ten added new value. That team was busy every day, and almost none of the work moved the business forward.
Many teams are in this spot. They are in catch-up mode, not value-creation mode. The roadmap says growth, and the commit history says repair. It happens slowly. Every rushed release leaves a little debt, the debt produces incidents, and the incidents eat the next sprint. Nobody decides to stop building. The defensive work just keeps arriving first.[1]
It is not only engineering. A customer experience team that walks every new customer through a half-built onboarding by hand is playing defense. So is an operations team reconciling the same spreadsheets every week because two systems never got connected, or a sales team rebuilding the same proposal from scratch because nobody owns the template. Each workaround feels like good service. Together they are the reason the team never gets to attack its market.
We saw a version of this in our own work. When we looked at where our coding agent sessions actually went, a lot of them were defense, which is part of why I wrote about moving coding agents off the laptop. Most of the AI conversation is still about offense: generating more code, more content, more outreach. I have argued that this is an offense-only problem and that AI on defense is where much of the near-term value sits. Agents are good at the defensive backlog. It is repetitive, well specified and easy to check.[2] That is the missing half of most agent strategies.
So run the count. Sort the last month of shipped work in each function into fixes and new value. If defense is most of it, give that backlog to agents running in the background, with a person approving the results, and hand the hours you get back to the people who know where the market is open. Then check the split again next month and see whether the new-value share went up.
Key takeaways
- When 90% of pull requests are fixes, the team is maintaining the business instead of growing it.
- Defense hides outside engineering too, in every manual workaround that props up a half-built process.
- Let agents absorb the defensive work so people can spend their time attacking the market.
FAQ
How can a leader tell if a team is stuck playing defense?
Sort the last month of shipped work into fixes and new value. In other functions, count the hours spent compensating for broken processes versus building new ones.
Should agents be pointed at offense or defense first?
Start them on the defensive backlog, because it is repetitive and well defined, and give the time it frees back to new work.
Related Essays
Agentic Defense: The Missing Half of the Equation
Put at least as much AI effort into securing, debugging, and testing as you put into building. Here is what that looks like.
Put AI on Defense, Not Just Offense
Most developers use AI only to write code. The real opportunity is using AI to secure, debug, and test—deploying equal firepower on defense.
The Offense-Only Problem with AI Coding
Most developers deploy AI almost exclusively to ship features faster. The asymmetry is the problem — not the AI itself.