FOR .NET TEAMS | AI AGENTS GENERATE PRODUCTION CODE
Your team uses AI tools. I help your way of working mature, while we ship real features together.
No training from a distance. Building along and showing how it's done.
Your developers are experimenting with GitHub Copilot, Claude Code and Cursor. But the output stays the same, only the tooling changed. The backlog grows, velocity doesn't. AI generates more code, so PRs get bigger and review pressure rises with them. The potential is there; the way of working just hasn't caught up yet.
You're not looking for a course. You're looking for someone who joins the sprint and shows how it's done.
Agentic coding is software development where AI agents take on whole tasks in your codebase: changing code, running tests, refactoring and processing feedback. A developer's work shifts from typing to specifying, reviewing and steering.
The gain isn't in faster typing, but in shorter lead times: less waiting, more parallel tasks, faster iterations. Provided the way of working around it is in order. If it isn't, you get 32 review rounds on a single PR and growing verification debt (articles in Dutch).
APPROACH
In three steps from experimenting to a way of working your team carries.
Together we map the concerns that live in the team and the technical challenges in your codebase. We only take first steps once the risks, objections and preconditions are clear and shared across the whole team, with real buy-in. Nothing gets rolled out over the team's head.
Two days a week in your sprint. In the first week we work on a real feature or improvement in your own codebase, not a standalone demo. From there we iterate and refine the way of working as we keep building.
The context files, agent instructions and review agreements then live in your normal way of working. I make myself redundant, and that's the whole point.
Safe and maintainable. That's what I steer on. AI fills missing specs with assumptions, and without review on architecture and security, mistakes slip in that only surface in production. I help build in extra checks and keep the judgement with a human. The tool speeds things up; you stay the owner, responsible together with your team.
People sometimes ask: where do you get this knowledge, where did you learn it? Not from a course. I build Invullen.nl entirely with Claude Code and maintain Factuur-Assist.nl (135+ customers) with it daily. So I know not just the demos, but the mess around them: failing tests, oversized diffs, odd assumptions and context that isn't applied. 20+ years in .NET at ABN AMRO, PGGM and Van Lanschot among others. Microsoft Certified: Azure AI Apps & Agents (AI-103) and Identity & Access (SC-300). A concrete example: I migrated the front-end of a legacy part of Factuur-Assist with agentic coding in 5 days, work I'd previously have set aside around 20 days for.
Honest about where this stands. I run agentic coding on my own SaaS every day, that's my workshop. As a standalone service for other teams it's newer, and that holds for everyone: this way of working has only been seriously usable since mid-2025, so nobody has a long string of engagements on it. The question isn't who's done it ten times, but who's deepest into it through real daily practice. That's exactly why I keep it small and testable: one team, one codebase, one real feature in week 1. If it doesn't work the way you hoped, you'll know after the first sprint, not after six months.
More about meFOR YOUR LEAD OR MANAGER
In short: I join two days a week and get agentic coding working in your own codebase. What it delivers: your team gets more done with the same people, at better quality, because agents can also help with testing, analysis and preparatory review, while the final judgement stays with the team. No quick fix: it's slower at the start, and the payoff comes once the way of working sticks and the team carries it. No dependency on me: your team stays the owner of the process and the guardrails.
One line to forward
"This .NET dev joins 2 days/week, gets our AI workflow actually working, and leaves the team able to run it themselves. First feature in week 1; tooling precondition (Copilot/Claude Code) and rate you discuss in an intro call."
Building software with AI agents that work autonomously in your codebase: generating code, running tests, refactoring. You specify, review and steer; the agent does the work. Tools like Claude Code, GitHub Copilot and Cursor make it possible. The difference with autocomplete is that the agent takes on whole tasks, not single lines.
With the right discipline, yes. AI fills missing specs with assumptions, and without review on architecture and security, mistakes slip in. I help set up review moments starting at the spec, guard security and maintainability, and keep the judgement with a human. That's exactly what I steer on. On top of that, the models themselves are improving fast at coding, Claude's and OpenAI's in particular, so the baseline quality keeps rising. What makes it safe for production is still the discipline around it, not the model alone. A good first step is usually improving test coverage, partly with AI itself; solid tests catch the mistakes that would otherwise only surface in production.
No. I work daily with Claude Code and I'm experienced with GitHub Copilot, and I also help with multi-provider setups. The approach is about the way of working, not one tool. Whichever AI provider your team uses, the principle stays the same.
I prefer to join two days a week, for a number of sprints. In the first week I want to ship a feature in your codebase, so the team sees live how it works. After that I phase out my role; the goal is that your team can continue on its own.
That's normal and often justified. Resistance is rarely about the tooling. We try to surface the unspoken concerns and bring the team along, instead of convincing with yet another demo. Those very objections can set the rules for the context and processes around AI.
Honestly: I run agentic coding daily on my own SaaS, but as a standalone service for teams it's newer. That's why I deliberately start small and testable: one team, one feature in the first week. You quickly see whether it works, and your team stays the owner of the process. What I have done for years: working inside teams as a senior .NET developer and architect at ABN AMRO, PGGM and Van Lanschot among others.
Fair question, and usually the first one I get. In short: the business plans of Claude (Team/Enterprise) and GitHub Copilot (Business/Enterprise) don't use your code to train models by default, that's built in, not a separate toggle. IP-sensitive parts of the code can also be excluded from AI. Does data have to provably stay within the EU? Then that drives the tool choice: Copilot offers EU data residency out of the box, Claude via an EU route on AWS or Google Cloud. We arrange that up front with your security team. Your team takes out the subscription with GitHub and/or Anthropic itself; I don't sit in between. My engagement runs through a standard contract for services. Working within security and compliance constraints is the norm for me; I come from the financial world (ABN AMRO, PGGM, Van Lanschot).
Let's get acquainted first, an hour-long call. I listen, ask about your stack and team, and give an honest assessment: where the gain is and whether I'm the right person.
Prefer to email first? Send your questions, and tell me briefly what your team uses now: Copilot, Claude Code, Cursor or something else.