Most development teams use AI blindly.
I help them in achieving a shared understanding with the agent, setting up appropriate boundaries for it and producing code that stays maintainable by people and machines. I apply classical software engineering practice to AI-augmented development.
Three failure modes
Getting the agent to understand
The agent builds something good. It's just not what you had in mind. Nothing is broken and the tests pass, but it answers a question nobody asked or builds the right feature in the wrong way.
I found it improves results when a shared understanding is reached between agent and developer before a line of code is written. When documented well, this understanding grounds and guides the agent during implementation.
AFK or Human-in-the-loop?
The agent plugs away at a big feature by itself. After 2 hours you find out it took a wrong turn 5 minutes in. Having learned from this, you watch its every move next time, approving every tool call. But it one-shots the whole thing, and you spent an hour watching.
I teach developers to judge which tasks an agent can be left to do, given a restricted environment and a good brief, and which need someone watching.
Maintainability for humans and agents
The agent writes code that is technically correct but difficult for humans to understand or modify. Humans then break a convention the agent invented, without knowing it was there, which in turn confuses the next agent. Inefficiencies like these limit what the team gets out of working with agents.
I document standards and conventions for the quality attributes that are key to the specific piece of software (e.g., code quality, maintainability, performance...) and set up short feedback loops around them, where new conventions are recorded as they are established and are easily referenced and enforced. Both humans and agents benefit: the next change is easier for whoever makes it.
What can I offer?
Training and coaching on using AI agents effectively
- How to establish a shared understanding between developer and agent
- How to avoid an agent doing unintended (or dangerous) things
- How to get humans and agents working together, doing what each is best at
- How to set up effective AI development workflows in practice
Fullstack engineering
- TypeScript and Node.js across the stack
- Frontends in all major frameworks (e.g. React and Vue)
- Infrastructure as code & DevOps practices
- Architecture and technical direction for agile teams
Background
I've spent 15+ years writing front- and backends on the web in small teams. I focus on building a process that encourages clean, maintainable code through automation of the boring parts. Whether that is linting, testing and deployment, or an environment and culture where agents are used well, it enables developers to concentrate on the hard and creative work, where real value is added.
Contact
