How I work, written down.
If a way of working isn't written down, it changes with every mood. This is the version I hold myself to, and the one my team can hold me to.
Five rules I don't negotiate.
- 01
Build first, plan second
The fastest way to the right idea is a rough version of the wrong one. Plans are better when they’re written about something real.
- 02
Make it safe to break
Trial and error only works inside a sandbox. Building the sandbox is the first job, not an afterthought.
- 03
The closest person decides
Technical calls belong to whoever is closest to the problem and will live with the result. My job is to make sure the right person is deciding.
- 04
Replace, don’t rewrite
Starting over feels like progress. Changing a running system piece by piece usually is progress.
- 05
Decide what not to build
Every yes has a cost someone else will pay later. Most of strategy is a well-defended no.
The decision goes to the closest person.
Pick a role to see what it owns. The closer you are to the work, the more of this table is yours.
| Decision | ||||
|---|---|---|---|---|
| How a system is builtThe person who’ll maintain it. | Decides | Consulted | Informed | — |
| Tools the team usesThey use them all day. | Consulted | Decides | Informed | — |
| Build, buy or integratePrototype first, then decide. | Consulted | Consulted | Decides | Consulted |
| When it shipsReadiness, not the calendar. | Consulted | Decides | Consulted | Informed |
| What AI agents may touchPermissions are a leadership call. | Consulted | Consulted | Decides | Informed |
| Budget and prioritiesBusiness sets the goal; we say what it costs. | Informed | Consulted | Consulted | Decides |
How to work with me.
What everyone who joins the team should know on their first day.
- How I decide
- I ask who is closest to the problem, then I back their call. I step in only when a decision reaches outside their area, or can’t be undone. When I overrule someone, I say why, in writing.
- Show me something
- A rough prototype beats a polished proposal. If you have an idea, build the smallest version that proves it, then come and argue for it.
- Bad news
- Bring it early and unpolished. A problem I hear about on day one is a conversation. On day thirty it’s a project.
- Disagreeing
- Do it in the room, with reasons. I’d rather be argued with than obeyed. Once we decide, we all commit, including me when I lost the argument.
- Meetings
- If it can be a message, it should be. If it has to be a meeting, it ends with a decision and a name next to it.
- What I expect
- Own the outcome, not just the task. Tell me what you’d do, not only what’s wrong.
- What I owe you
- Context, cover, and the budget fight. You shouldn’t have to argue with finance to do your job.
- Mistakes
- Mistakes made in the sandbox are tuition. Mistakes hidden from the team are the only kind I take personally.
The ground I work on.
Languages, platforms and tools I've built with, run, or led teams on.
Languages
- TypeScript
- JavaScript
- Python
- C#
- Go
- SQL
- PHP
Web & frontend
- React
- Next.js
- Astro
- Vue
- Tailwind CSS
Backend
- Node.js
- .NET
- FastAPI
- Django
- Laravel
Data
- PostgreSQL
- SQL Server
- MySQL
- Redis
- MongoDB
- Elasticsearch
Cloud & infrastructure
- AWS
- Azure
- Google Cloud
- Cloudflare
- Linux
- Nginx
- Docker
- Kubernetes
- Terraform
Delivery
- Git
- GitHub
- GitHub Actions
- GitLab
Integration & ERP
- SAP
- Odoo
- Dynamics 365
- Oracle
- Kafka
- RabbitMQ
- n8n
Observability
- Grafana
- Prometheus
- Sentry
AI
- Claude
- Claude Code
- OpenAI
- Cursor
- Ollama
- Hugging Face
Team & product
- Jira
- Linear
- Notion
- Slack
- Figma
What I kept, and what I didn't.
Kept, dropped, overrated, or still trying. Every verdict comes from real use, not a launch post.
Daily
- VS Codekept
Boring, fast, everywhere. Boring is a feature.
- Claude Codekept
Most of my building starts here now. I review everything it writes.
- Git + GitHubkept
If it isn’t in version control, it didn’t happen.
- Cloudflare Workerskept
This site runs on it. Push to main, live in a minute.
Verdicts
- Agent loops & scheduled agentstrying
Promising for repetitive work. Still learning where “done” is.
- AI meeting summariesoverrated
Accurate about what was said. Useless about what was decided.
- Heavyweight project boardsdropped
We spent more time grooming the board than shipping.
- Low-code for core processesdropped
Fast to start, expensive to escape.