38 percent AI code, 22 percent more defects: What AI assistants cost your engineering team
AI assistants were supposed to make engineering faster. More code, less routine, happier developers. License numbers go up, the surveys are glowing and the board considers the topic done.
Then someone opens the ticket system.
Almost four in ten lines of new code come from the assistant. In the same quarter, defects that customers find first rise by more than a fifth. The time saved while writing is lost again in support, and nobody measured it.
License counts and surveys measure how much a tool is used. What it costs in operations shows up in review times and incidents, and hardly anyone looks there.
This keynote shows which tasks assistants speed up and where they create defects that surface months later. It also covers two rules that bring the defect rate down without giving up the assistants.
You leave with:
- - four metrics you can pull from your own ticket system within a week
- - two rules that catch defects from AI-written code
- - a solid answer when your board asks what the AI licenses are worth
Writing faster is easy. Shipping faster is a different story.
- Format
- Keynote
- Length
- 40 minutes
- Audience
- CTOs, heads of engineering and engineering leads
What the audience takes away
- Four metrics that teams can pull from their own ticket system within a week
- Two rules that brought the defect rate down again by December
- A year of measurements as a basis for the next decision on AI licenses
Matching stages
- AI in Engineering Summit
- Systems Reliability Conference
- Engineering Leadership Days