90% AI Adoption Across 13 R&D Teams

Challenge

Heureka Group's 100+ person R&D organization was already experimenting with AI, but unevenly. Some teams had power users with advanced setups. Others hadn't started at all. There was no shared playbook, no governance, and growing uncertainty around security and costs.

The risk was clear: without structure, isolated experiments would never scale. Power users would keep advancing alone while most teams stayed stuck at basic usage or avoided AI altogether.

Key blockers we found across teams:

• Security concerns about code leaving the organization.
• Budget and license caps limiting tool access.
• No shared knowledge of context management techniques.
• Knowledge silos between teams solving similar problems.
• Dev-focused tools that didn't fit Product and UX workflows.

Our Solution

We embedded directly with Heureka's teams for three months in Q4 2025, working alongside them rather than advising from outside.

First, we mapped the territory. 13 in-depth interviews with every major R&D team. We classified each into adoption levels (L0 through L3) based on actual behaviors, not self-reporting. This gave us a clear picture of where to invest effort.

Then we built the engine. We formalized the AI Ambassador program and expanded it 3× by bringing in people from Dev, Product, and UX. We delivered 8 Tech Talks, supported hackathons, and created practical guides that teams could use the same day.

Then we proved it works through pilots:

• A specialized documentation agent "interviewed" a legacy PHP monolith and produced accurate, nested documentation. Work that would take a senior developer weeks was done in hours.
• We built a Claude Code integration for GitLab CI/CD that pre-reviews every Merge Request, catching security gaps and naming inconsistencies before a human reviewer looks at it.
• We connected AI tools to Jira, Confluence, GitLab, Sentry, and live library docs through MCP, giving agents full organizational context.

Everything we built went into a shared knowledge base: the Agentic Blueprint repository (clone-ready), AGENTS.md templates, MCP setup guides, and a security-vetted tools list. All running under company licenses with contractual guarantees that code stays private.

Results

Three months in, the numbers tell a clear story:

• 90%+ of R&D now regularly uses AI-assisted tools.
• 50%+ of individuals use AI in their daily work.
• Teams report 50% time savings on certain tasks, with some achieving 2–3× speedup.
• Legacy documentation generated in hours instead of weeks.
• AI Ambassadors recognized as the go-to resource across the organization.

The biggest shift isn't in the metrics. It's in how teams think about AI. Teams that started at Level 0 now have tools, a knowledge base, and direct support. Teams at Level 2 are building PoCs that wouldn't exist without AI and iterating faster because rewriting is cheap.

Heureka's engineering team published a detailed look at the AI Ambassador program on the HeurekaDevs blog. As part of the collaboration, our consultant even joined Heureka's internal hackathon and won first place with the team, building ASMA (AI Strategic Merchant Advisor) for automating merchant strategies.

The Agentic Blueprint, templates, and guides enable self-service adoption. The progress continues without us.