Case study · Ucopia
Ucopia: AI in the development of 20-year-old software, without compromising on robustness
Gérald Colangelo, Head of R&D - CISO at Ucopia, wanted his development team to deliver faster with AI, without compromising the product's high standards. Mission accomplished.
- Software
- Ucopia, created in 2002
- Deployments
- Over 12,000, B2B
- Team
- Around ten developers
- Engagement
- 5 days
Gérald ColangeloHead of R&D - CISO, UcopiaUcopia“To me, AI is a real opportunity: moving faster, meeting our customers' needs better, and handling tedious work such as migrations, test plans or documentation more efficiently. But it is also a source of risk. On a product like Ucopia, where security and robustness come first, it is only worth it if we roll it out in a controlled way, adapting how we work. We are the experts on our product. Ask This Guy knows AI technologies, their gains and their limits, and quickly helped us identify what was suitable and genuinely useful. Each to their own expertise.”
Where we started
Ucopia manages Wifi access for professional sites: captive portal, authentication, filtering. Created in 2002, it has over 12,000 deployments, from businesses to airports, hospitals and stadiums. Security, stability and robustness are non-negotiable, and twenty years of history have produced a rich codebase, written in many languages, where every change is delicate.
Around ten developers work in different editors (Vim, IntelliJ, PyCharm, VS Code). A developer is more productive in a familiar environment: AI had to fit into it, not replace it. Model usage also had to stay under control.
How AI entered Ucopia's development
Understand the team
Interviews about practices, development environments and shared rules. Three priority uses: querying the code, reviewing merge requests, auditing the legacy code.
Choose tools and models
The AI features available in each editor, European and US inference providers, code-specialized models: chosen on cost, performance and control over usage.
Shared rules, in everyone's editor
One central instructions file (conventions, architecture) that every editor reads. Each developer keeps a familiar environment, and the AI follows the same rules there for the whole team.
Deliver quick wins
Regression checks, documentation and architecture diagrams, file-by-file legacy audit: visible gains from the first days, on Ucopia's code.
What AI changes for developers
Deliver faster
Developers query the code in plain language and find the right place to change. READMEs, project structure and architecture diagrams are generated from a single prompt, which speeds up getting to grips with a module.
Better code quality
A file-by-file audit of the legacy code, with an open-weights model, surfaces the fixes to prioritize: mishandled exceptions, infinite loop risks, unreleased resources. AI-assisted review complements the existing linters.
Robustness preserved
Nothing goes through without human review. Automated regression checks have already exposed a bug that had gone unnoticed, and shared rules keep the product's conventions whatever the tool.
With AI, the developer matters even more than before: they become the conductor. They set the scope, give the instructions, review and decide.
Mission deliverables
Map of the team's tools
The AI features available in each environment (Vim, JetBrains, VS Code, VSCodium, Cursor) and command-line agents, with their limits.
Providers and models comparison
European (Scaleway, Nebius, Mistral) and US inference providers, code-specialized models, compared on price, per-project quotas and performance, plus a proxy to track and cap usage.
Assessment on the code and recommendations
What adds value on the Ucopia codebase, what to avoid, and the order in which to roll it out to the team.
Does your team want to adopt AI without losing quality?
We start from your tools and constraints, then put AI to work on your code. You keep what works.