Posts
Beta Noise Is Fine. Not Knowing Which Noise Matters Isn’t.
Running agentic features in beta means expecting errors. The problem we had wasn’t the errors — it was having no way to tell which ones needed urgent action and which ones could wait for the next sprint.
Four Font Sizes, Not Five
AI makes design decisions fast and invisibly. You can’t audit them by eye the way you’d inspect your own work. So I set up a step where AI audits itself — before any code runs.
Two Questions Before You Let Claude Start
Before letting Claude write a single file, ask which account it’s signed in as and which working directory it’ll write to. Two PM-style checks most sessions skip.
My Accounting Notebooks, Now Inside Claude
I’m connecting NotebookLM to Claude via MCP so questions about Swedish tax, migration, and social insurance get answers grounded in actual Skatteverket and Migrationsverket guidance — not just training data.
Why I Use a Framework for AI Coding Now
A single AI agent can’t verify its own plans. A framework puts a second pair of eyes on the quiet drift you’d miss on a skim.
Writing Docs for a Collaborator Who Forgets
Writing docs for AI coding assistants isn’t the same as writing them for people. It needs a reference doc in the repo, a code anchor, and a commit.
Small World
How building the assistant side of a care app revealed a product-led growth mechanism already built into how the care community works — and why I almost missed it.
How to Run a Backlog Refinement With Claude
I built the guardian side of this app without running a proper backlog refinement. The assistant side is where I tried to get it right.
Day Zero
How to plan a vibe coding project before writing a line of code — prompts as product briefs, scope decisions logs, and why GitHub Issues are model-agnostic context.
Claude Will Build the Whole Thing If You Let It
AI tools optimise for completeness. They don’t have a launch date, a budget, or an MVP. The question ‘is this overkill?’ is free to ask — and it’s the one that keeps your v0.1 the right size.
The Compliance Rules My App Was About to Ignore
I thought I needed a Claude Skill to store the FK rules. I was wrong — and learning the difference between Skills, memory, and reference documents changed how I think about building with AI.
The Agent I Didn’t Plan For
I always planned to build AI agents — for sales, marketing, support. I didn’t expect the first one to be a compliance expert. And I didn’t expect it to arrive before the app had a single user.
How to Run a Sprint Planning Session with Claude
I run sprint planning sessions for a living. One session with Claude cut a feature, rewrote two user stories, and uncovered a gap I never formally declared out of scope.
The Instruction Sheet I Wish I’d Written From Day One
Claude calls them Skills. ChatGPT has Custom Instructions. Gemini has Gems. Every major AI tool has a version of this — and most people haven’t touched it yet.
I’ve Written These for Years. Never for My Own.
I use AI to write user stories at work. I skipped them for my own app. The basics matter more when there’s no team to catch what you miss.
I Told It to Tell Me What It Learned
One instruction made my AI custom instructions smarter every session. The pro tip nobody mentions when setting up Claude Skills or ChatGPT Custom Instructions.
How I Turned Acceptance Tests Into a Regression System
I ran my first acceptance test on the Kalinga app. Sixteen features, two bugs found in five minutes, and two user stories that described an app that no longer existed.
Your Network Is a Vibe Coding Tool. Most People Forget to Use It.
AI has limits. When Claude installed Python in my Node.js project, I turned to my network. Why the people around you are an underrated vibe coding tool.
Why I’m Writing This
I don’t really post. But I documented this vibe coding journey because my network is ahead of me on AI — and I wanted them to see it’s possible without being technical.
Debugging a Vibe Coded App: Silent Bugs, Wrong Field Names, and a Missing Password
I locked myself out of my own app. Then found silent bugs on every page. Debugging a vibe coded app: wrong field names, missed FK deadline, blank dashboard.
How I Decided Which Features to Cut from My Vibe Coded App
Removing features from a vibe coded app felt like clarity, not loss. How I applied an MVP test to decide what to keep and what to cut.
I Built Everything I Could Think Of. Then I Cut Most of It.
When vibe coding, feature creep happens when excitement outpaces validation. How MVP thinking helped me cut back a vibe coded app that did too much.
Vibe Coding Mistake: Why I Built Wireframes Before Reading the Requirements
The first vibe coding mistake I made: building wireframes before reading the requirements. Weeks of work before I solved the actual problem.
I Spent Years Correcting the Same Forms. Then I Vibe Coded an App.
A non-developer managing personal assistance paperwork in Sweden. After years of errors filling FK forms, I vibe coded an app to fix it.