Atlassian and 1Password Just Killed Their AI Centers of Excellence

Atlassian's AI center of excellence lasted exactly 18 months. The company dissolved it last week, reassigning all 47 engineers to product teams across Jira, Confluence, and Trello. 1Password made the same move in March, embedding their central AI team directly into security and authentication squads.

The timing isn't coincidental. Both companies cite the same Q1 2026 productivity data: teams with embedded AI engineers ship features 2.5x faster than those relying on a centralized AI unit.

The Center of Excellence Model Breaks Down

Centers of excellence made sense in 2023. Companies needed to move fast on AI, and centralizing expertise seemed logical. Build one team of specialists. Let them develop best practices. Have product teams queue up for their help.

That model worked when AI was an experiment. It fails when AI becomes core to every feature.

"We had teams waiting six weeks for AI engineering support," said Avani Prabhakar, who led Atlassian's transition to the embedded model before taking on a new executive role focused on internal AI adoption. The bottleneck wasn't just timeline. Centralized AI engineers lacked product context. They'd build technically sound features that missed user needs.

Microsoft's Early Warning

Microsoft saw this coming. At DX Annual in May, CVP Tim Bozarth shared data from their own transition: "Embedded AI engineers don't just code faster. They make better product decisions because they sit in daily standups, hear customer feedback directly, own the full stack."

The numbers back this up. Microsoft's embedded teams reduced AI feature development time from 12 weeks to 4. More importantly, user adoption of AI features jumped 3x when built by embedded engineers versus centralized teams.

What Embedded Actually Means

1Password's approach offers a blueprint. When they refactored their multi-million-line Go monolith using AI agents, they didn't hand the project to a central team. Instead, they embedded AI engineers directly with the teams who owned that code.

Nancy Wang, who co-led the refactoring project, explained their logic: "AI agents succeed when they have deep context. That context lives with the engineers who wrote the original code, not in a center of excellence three org charts away."

The embedded model doesn't mean every team gets a dedicated AI engineer. At Atlassian, most squads now include one AI specialist who splits time between 2-3 related teams. These engineers report to product leadership, not a central AI organization.

The Talent Redistribution

Dissolving centers of excellence creates its own challenges. Where do you put 47 AI engineers? How do you maintain technical standards without central oversight?

Atlassian solved the first problem through what they call "AI pods." Each major product area (Jira, Confluence, Trello) now has 8-12 embedded AI engineers who meet weekly to share learnings. They maintain technical consistency without the bureaucracy of a central team.

The second challenge proved easier than expected. "Standards emerge naturally when engineers own outcomes," Prabhakar noted. "Central teams create standards for compliance. Embedded teams create standards because bad AI reflects directly on their product metrics."

The 1600-Person Context

Atlassian's shift to embedded AI comes weeks after announcing 1600 redundancies. While the company hasn't directly linked the layoffs to their AI restructuring, the timing suggests a broader rethinking of team composition.

Traditional engineering roles are shrinking. AI-augmented roles are expanding. The embedded model accelerates this shift by putting AI capabilities directly where product decisions happen.

What This Means for AI Hiring

The death of AI centers of excellence changes how companies recruit AI talent. Job postings increasingly emphasize product sense alongside technical skills. "Can you train a model?" matters less than "Can you identify where AI improves user experience?"

Salary data reflects this shift. Embedded AI engineers at Atlassian and 1Password earn 15-20% more than their centralized counterparts did. The premium isn't for technical skills. It's for the ability to bridge AI capabilities with product needs.

Companies still experimenting with AI will likely maintain centers of excellence through 2026. But for those where AI drives core features, the embedded model isn't just more efficient. It's the only model that scales.