Frederic ZINGG — French/Swiss Master's in Mathematics & Computer Science

Most agent projects do not fail on the model. They fail because nobody established how the work really runs before automating it, or because the agent cannot reach the systems where that work actually happens.

That gap is the job. Companies come to us with a process or a product they want to make agentic, and what they arrive with is rarely the problem statement they need. Closing the distance is half business, half engineering — and the business half decides whether the engineering was worth doing.

On the business side: time with the people who own the work, and a brief that survives contact with a builder. Designing what an agent does — policy, plans, skills, knowledge — is a separate discipline from choosing its tools, and the two only work co-designed. The behaviour-design method our engagements run on comes out of that: four phases, each gated, and a repertoire derived from real user utterances pulled out of call transcripts and support tickets. Skills invented top-down from plausible-sounding categories are the most common reason an agent feels wrong. Engagements close with a handover package: a standalone, documented repository the customer's own team runs and extends without us.

On the engineering side: spec-first development with agents. Acceptance criteria before code, agents implementing against them, verification that proves the system does what the spec claimed — not a demo that happened to work. Core contributor to Memex (memex.ai), a source-available specification-and-verification platform for AI-native teams, across its API server, React UI, acceptance-criteria emission library and deploy pipeline. Also the delivery platform my own team runs on, built end to end, and the architecture of an agent-first platform rebuild where every capability is reachable identically by a UI and by an agent over MCP.

The connective work matters more than either half alone: MCP servers, API services and integrations that let agents act inside learning and workforce systems, compliance tooling, developer portals, internal APIs and data warehouses.

30 years shipping software underneath it — engineer's degree in computer science and applied mathematics; C++, Rust, Ruby, TypeScript, Python, Java, Go; dozens of REST APIs; cloud-native data platforms on Google Cloud. Infrastructure held to the same standard as application code.

Working at Mindset.ai

• Contributed to Hyperstack, Ruby-based DSL and modern web toolkit for building spectacular, interactive web applications fast!
• Contributed to Intekt / IEML, a language for metadata designed to structure and categorize information on the Internet.