AI Strategy6 min
How to tell if your business actually needs AI
Most AI projects fail during selection, not execution. A short test for whether the problem in front of you is one a model should be solving.
Insights
Notes from building the thing.
Writing on AI product development, engineering and the specific ways projects fail between a working prototype and a working product.
AI Engineering8 min
Why AI prototypes fail when they reach production
The demo worked on every example you tried. Here is the specific set of things that breaks between there and real users, in the order they usually break.
App Rescue7 min
Your AI app works locally. Now what?
A practical checklist for the gap between 'runs on my machine' and 'safe to put in front of users' — in priority order.
App Rescue7 min
How to rescue a half-built application
The instinct is to rewrite. It's usually wrong. A method for assessing what you have and deciding what to actually do with it.
AI Product Development6 min
Building an AI MVP without overengineering it
Which parts of the 'proper' architecture you genuinely need on day one, and which are expensive ways to delay finding out whether anyone wants this.
Build With AI8 min
From Cursor prototype to production application
AI coding tools get you to a working prototype remarkably fast. What they consistently leave out, and how to add it back deliberately.
In progress
Being written.
Outlined and queued. Listed here rather than published as empty pages — these aren't live yet.
- AI EngineeringDraft
The smallest useful AI evaluation harness
You don't need an evals platform to know if a prompt change helped. A fixed test set, a scoring function, and a number — the minimum that actually works.
- App RescueDraft
Why your first App Store submission keeps getting rejected
The rejection reasons that catch first-time submitters every time, and the checklist that would have caught them before Apple did.
- App RescueDraft
You inherited a codebase and the developer is gone
No documentation, no handover, no one to ask. A method for making an unfamiliar codebase safe to change before you touch a single line.
- AI Product DevelopmentDraft
What an AI MVP actually costs to run
Model calls are the line item nobody models until the first invoice. A framework for estimating per-user cost before you launch, not after.
- AI StrategyDraft
Build AI in-house, hire a studio, or buy the tool
Three real options, three different failure modes. A way to reason about which one fits your team's actual constraints, not its ambitions.
Beyond reading
Rather have us look at yours?
The writing is general. The engagement isn't.