Thoughts on software architecture, startup building, and lessons learned from shipping products.
There is a moment in every project where the AI agent tells you something that does not match what you can see. The answer sounds confident. The disclaimer says always double-check. Most people do not. This is a story about the third time I pushed back, and what I learned about when to override an agent that sounds right.
Two hundred markdown notes nobody read. Six projects in flight. A million-token context window that still ran out. The Founder Knowledge Graph started because the way I was working with AI agents was breaking down. Here is what I learned, and what I built.
Some parts of software run whether you ask for them or not. Some only run when you know to ask. AI app builders are mostly the second kind, and the slogan never names the difference. A look at what you are actually buying when you pay for 'no developer needed.'
You can test every feature twenty times in dev and production will still find something you missed. This is the story of the night I learned that, on my own, at two in the morning, with nobody else to call.
I have spent years in marketing. I know the tools, I know the patterns. And I am going to say something most people in this industry will not say out loud: marketing in 2026 is a black box. Here is what has changed and why I do not have a clean answer at the end.
Most founder content is either 'I quit my job and went all-in' or 'I am still at my day job and have no time.' Neither describes what I actually do. Here is the honest middle path, and what running two tracks at once has taught me about building and shipping.
Most SaaS products start with one dashboard and different views per role. I did not do that. Here is why separate dashboards for homeowners, contractors, and admins was the only honest answer, and why there is a fourth dashboard I have deliberately not activated yet.
The tools make parallel projects possible. They do not make them easy. Here is what running five projects solo actually looks like, and what it costs in the parts of your life you do not read about on social media.
Readers kept asking me how RLS actually works, when to use it, and how to avoid tanking performance. Here is the long answer, based on what I have learned building Havnwright on Supabase.
Writing rules for your AI agent gets you to about forty percent. Infrastructure takes you to eighty or ninety. Here is why, and what to do about it.
The articles on Havnwright are not a launch strategy. They are seven months of writing what I was researching anyway. Here is how I think about content as a solo founder, and why most of the standard advice is wrong.
The building was three weeks. The submitting was three weeks. If you think the hard part is writing the code, you have not tried to get an app through the Apple and Google review process yet.
Most people talk about AI memory as a feature. The real problem is not features. It is ownership. This is the story of why I stopped waiting for built-in memory and built my own.
Not the demo-day version. Not the viral tweet version. What building solo with AI actually looks like, day after day, when you are trying to ship something real.
Managing a four-month renovation with nothing but receipts and phone calls taught me that the industry needed a system. This is the story of how that experience became Havnwright.
Why your auth keeps breaking in production, and the battle-tested architecture that companies like Vercel, Stripe, and Notion use to prevent it.
From playing games on my computer at eight years old to building products people actually use. This is the story of how I got here, and why it took so long.