Your engineers use AI to write more code. Where's the AI to fix it?

An SRE said this to me in a conversation, and it has stuck with me since: "The engineers now have AI to write code, and they're writing it three times faster than last year. Why don't I get any AI to fix the bugs all this code has generated?"

It's a fair question. And the answer is that most of the industry built AI for one half of software and left the other half where it was.

The writing side got a decade of tools in a year

Google's 2025 DORA report found that 90% of technology professionals now use AI at work, and more than 80% say it has made them more productive. With code assistants and agents, the speed at which code gets written has changed more in eighteen months than in the decade before.

The same research found something less celebrated. Higher AI adoption is associated with an increase in both delivery throughput and delivery instability. More code ships, and more of it breaks.

That second finding lands on one group of people: the ones carrying the pager.

Fixing is a different kind of work

Writing code starts from intent. You know what you want; the AI helps you get there faster. Fixing an incident starts from a symptom. Something is wrong, you don't know what, and the clock is running. The work is investigation: what changed, where, and why the thing that alerted may not be the thing that broke.

That's why toil hasn't fallen the way people expected. Catchpoint's SRE Report 2025 found median time spent on toil rose from 25% to 30%, the first increase in five years. Their own researchers said the expectation had been that AI would reduce toil, not add to it.

The tools pointed at incidents mostly did one of two things. They summarised the alert you'd already read. Or they promised autonomy that no responsible team would switch on in production.

What the fix side actually needs

If you design AI for the people who keep production up, three things matter more than anything else.

It has to know your systems, not the internet.
A general model has read the public web and none of your incidents. Most of what breaks has broken before, in a way specific to your architecture. DORA's own guidance is to connect AI directly to your internal data, because generic tools struggle with the last mile of real systems.

It has to show its work.
DORA found 30% of developers have little or no trust in AI-generated code, partly because AI tools often can't signal when they're uncertain. An engineer at 2am can't afford a confident wrong answer. They need a confidence level and the evidence behind it, so they can check it in seconds.

It has to leave the decision with the engineer.
When something goes wrong in production, a human name is on it. Good help does the legwork around the decision: gathering context, correlating changes, keeping people updated. The engineer makes the final decision.

This isn't about replacing anyone

There's a version of this story where AI makes SREs unnecessary. We don't believe it, and the people doing the job don't either.

Reliability is the hardest problem in software, and it's getting harder. Uptime Institute expects failures to increasingly come from complex interactions between systems rather than a single point of failure. That's exactly the kind of problem that needs experienced human judgment. What it doesn't need is that judgment spent opening twelve tabs and searching Slack for a thread from last March.

The goal is to give your best engineers their time back: less scrambling, more building systems that don't fall over.

The other half

That's why we built Sera, the AI SRE teammate inside Xurrent IMR. It works the incident alongside your team: investigating the likely cause from your alerts, incident history and recent code changes, keeping stakeholders updated, answering questions, and drafting the postmortem. Your engineers make every decision.

Your engineers already have their AI. Your on-call team deserves theirs.

Sources: DORA, Balancing AI tensions, March 2026 · Catchpoint SRE Report 2025 · Uptime Institute Annual Outage Analysis 2026

‍