A developer discovered that AI agents replicate the same ticket fragmentation problems that have plagued human teams for decades. When given narrow, technical tickets, agents immediately break work into atomic issues, pushing tasks between tickets rather than solving problems holistically.

In a controlled experiment, an agent wrote a clean, professional ticket describing a bug and proposed fix. The next agent followed this biased path exactly, introducing two new bugs because the narrow scope excluded critical context. This pattern repeated across three iterations, with the original outcome buried under fragmented fixes that collectively solved nothing.

The issue isn't agent behavior but the ticket patterns themselves. Agents behave like the tickets they learned from - scope-narrowing, deferring to small pieces over whole outcomes, fragmenting before thinking. These behavioral patterns were always present in human teams but hidden by hallway conversations and tribal knowledge that filled gaps. Without those informal channels, agents rely solely on ticket text.

The solution requires writing tickets as outcome summaries rather than technical tasks. Assign agents the biggest justifiable piece - a two-line product outcome or feature description. Let agents determine subtasks when work is ready for review, not upfront. Break large initiatives into smaller initiatives, not smaller issues. Each ticket should give agents room to reason, not a technical path to follow blindly.