Atlassian's Jira team migrates 100+ GraphQL fetchers off a legacy gateway in a week, about two quarters of work, using agents driven by Jira specs
Five Jira developers moved more than 100 GraphQL fetchers from a legacy gateway service to Atlassian's standard platform layer. AI first wrote a behavioural spec for each fetcher from the existing code (schema, upstream sources, transformations, business rules, acceptance criteria), engineers reviewed and corrected each spec in a Jira ticket, and agents then generated the migration code in parallel batches. Verification ran in three layers: acceptance criteria in the ticket, a separate validation agent that tested local code against the live service and flagged differences, and a shadow-traffic framework that forwarded requests from the old service to the new one and compared responses.
The numbers (as reported)
| Scope | 100+ GraphQL fetchers |
|---|---|
| Team | 5 developers |
| Time | "around 2 quarters worth of work in a week" |
Facts
- Company / subject
- Atlassian
- Tools
- Unstated
- Codebase
- brownfield
- Work type
- Migration
- Stack
- GraphQL
- Published
- 2026-06-25
- Last checked
- 2026-10-22
What the source claims Practitioner (own blog/paper)
Claimed by the builder, writing about their own work. Not checked independently.
- Five developers did around two quarters of work in a week, moving 100+ GraphQL fetchers ~2 quarters in 1 week
“We managed to do this with 5 developers completing around 2 quarters worth of work in a week”
Caveats
Self-reported on Atlassian's own blog, and the post doubles as a pitch for Jira as agent infrastructure. It names no agent, model or vendor, and gives no line counts, cost, defect or rollback numbers. The two-quarters figure is the team's estimate of manual effort.
More stories by tool: Unstated
Related topics: MigrationsBrownfield codebases