Johan Lajili

Johan Lajili is a software engineer at Poolside building tools that help AI coding agents inspect running applications, reproduce bugs, and verify their work. His focus on agent feedback loops draws on experience developing games, real-time sports infrastructure, enterprise document software, and AI-powered spreadsheets.

From games to production AI

Lajili studied game design and programming at Isart Digital in Paris, graduating in 2013. At Gamesys, he developed HTML5 games and a team game engine using reactive and functional programming. He subsequently supported CMC Markets’ transition from Angular to React before joining IMG Arena in 2019.

At IMG Arena, he built products delivering live statistics for tennis, golf, and mixed martial arts, consolidating real-time WebSocket feeds into a GraphQL subscription service. To scale that infrastructure without overwhelming upstream data providers, he designed a primary-and-secondary server architecture using Redis Pub/Sub. He became head of technology for web applications in 2021 and senior staff software engineer in 2023.

At Pactio and its Rowan AI initiative, Lajili built document tracked-changes functionality, helped develop multistep agentic workflows, and created testing infrastructure that measured reliability and latency. His AI-powered spreadsheet reconciliation combined conventional matching algorithms with language models that specified constraints such as date windows and acceptable differences between amounts. After Rowan AI’s codebase was acquired, he consulted for YLookup, restoring application infrastructure and improving PDF-to-spreadsheet testing and extraction. His engineering portfolio documents these projects; his React boilerplate repository reflects his earlier JavaScript background.

Making coding agents reliable

At Poolside, Lajili created Spoolside, an internal command-line tool that lets coding agents interact with the company’s VS Code extension. It supplies screenshots, token-efficient application snapshots, frontend and backend logs, service restarts, and controls for navigating menus and testing conversations.

His work on coding-agent verification centers on four practical convictions:

• Feedback determines performance in existing codebases. Legacy applications defeat agents when obsolete code paths, unfamiliar dependencies, and hidden runtime behavior prevent them from checking their assumptions.
• Reproduce the bug before proposing the fix. Agents should demonstrate the original failure and inspect the result, reducing the manual verification required from engineers.
• Build product-specific agent interfaces. A VS Code extension, permissions-heavy application, or game needs different inspection and interaction primitives, delivered through command-line tools, skills, or MCP integrations.
• AIX engineering makes autonomy trustworthy. Better application observability, clearer codebases, accessible knowledge, and stronger verification help prevent faster generation from multiplying errors.