We’ve all been there: A “simple” legacy migration is estimated at 6 months. Two months in, the team discovers a web of hidden dependencies and implicit logic. The timeline balloons from 26 weeks to 70. The business loses faith.
At the latest Tech Rocks Summit (Dec 2025), Quentin Pleplé (CTO at Theodo) shared a framework to break this cycle. By merging Lean manufacturing principles with Agentic AI, his team is now modernizing legacy systems 3x faster while actually increasing code quality.
The Strategy: The “Modernization Factory”
Instead of treating migration as a manual craft, Theodo treats it as a structured assembly line. The unit of work is the “Touchpoint”—any testable element like an API endpoint, a frontend page, or a cron job.
By moving from simple workflows to autonomous agents in 2025, they’ve defined a Migration Flow that ensures every piece of code is scrutinized by both AI and humans.
The 6-Step Industrial Flow Based on the process Quentin detailed, every touchpoint follows a rigorous path to ensure nothing is missed:
- Discovery (Agent-Led): An agent maps dependencies using recursive analysis to find “implicit” logic—hidden framework hooks or naming conventions—that traditional tools miss.
- Reverse Engineering: This agent extracts business rules and technical logic, converting “spaghetti” into structured Markdown documentation so the “why” is never lost.
- The Plan (Architect Agent): Using Spec-Driven Development, the agent writes the specs. Crucially, it must ask the developer 3–10 clarifying questions before it is allowed to proceed.
- Code & Test: Developers work in Pair Programming with the AI. A QE agent runs unit and integration tests in a loop until everything passes—but humans must validate every step.
- Kaizen (Continuous Improvement): Every human comment on a Pull Request (PR) is treated as a “defect”. A Kaizen Agent analyzes these via the GitHub API to find root causes and update the project’s global guidelines.
- Validation: The final human review ensures the architecture meets the target standards.
Engineering Rigor: Solving the “Reviewer’s Nightmare”
One of the biggest risks with AI is the “PR dump”—6,000 lines of code that are impossible to review. Theodo solves this with the Storyteller Agent.
The Storyteller Agent rewrites the Git history into a logical narrative. It breaks a massive PR into small, atomic commits (Documentation → Pure Logic → Services → UI), making it digestible for a human Tech Lead.
The Result: 21 Weeks Instead of 70
By applying this Lean mindset, HealthHero saw their “runaway” 70-week migration completed in just 21 weeks. The speed didn’t come from cutting corners, but from automating “no-value” tasks like boilerplate and test writing, allowing humans to focus on high-level architecture and refactoring.
Coming Soon: Le Podcast on Emerging Leadership
If this “Industrialized Agile” approach sparked your curiosity, you won’t want to miss our upcoming episode of Le Podcast.
We are sitting down with a renowned book author who will dive even deeper into how to effectively combine Agile and Lean methodologies within large-scale tech organizations. We’ll explore how to move beyond individual team “hacks” to create a true culture of continuous improvement across thousands of engineers.
Don’t miss out: Subscribe to Le Podcast on Emerging Leadership on your favorite platform to be notified the moment this deep-dive episode drops.
A Final Thought to Carry With You As we move from “handcrafted” code to the “AI Factory,” the most critical question for leadership isn’t how fast we can code, but:
In an era where agents can handle the implementation, is your team currently organized to be the architects of the process, or are they still just the builders in the trenches?
Leave a Reply