Tactical writing for founders building AI products. What we learn shipping marketing, healthtech, and fintech systems, written down while it is fresh.
Most advice on choosing a development partner is written by development partners, which is why it always concludes that you should look for communication, transparency and cultural fit. Nobody has eve
Every software company sells AI now. The demos are uniformly impressive, because a demo is the one environment where the model always behaves, the data is clean and nobody has tried anything unexpecte
Why we stopped selling fixed-price MVPs A fixed price sounds like the client carrying less risk. It is usually the opposite. What a fixed price actually does is move every disagreement about scope fro
The messages got shorter, then slower, then stopped. Maybe there is an unanswered invoice, or a promise about "next week" that has now been three. Either way you have a product, or most of one, and th
AI-built apps break first at authorisation, database access rules, leaked secrets and unverified payment webhooks, not at the feature they were built to demo. They fail at the thing nobody demonstrate
A code audit report earns its fee when it contains four things: an executive summary the person paying can act on, findings ranked by severity that cite specific files and lines, an architecture asses
Technical due diligence rarely kills a round. It re-prices one, and it does so at the worst possible moment: after the term sheet exists, when the fund has already decided it wants in and you have alr
A code audit for an early-stage startup costs somewhere between $49 and $15,000. That spread is not vendor margin. It is how many hours a senior engineer spends inside your system, and whether a perso
Comparing AI models on vibes is how products end up 10x over budget. The comparison that holds up is cost per completed task, latency at your context size, and behaviour on your own eval cases, not le
Choosing an AI development agency comes down to five checks: production evidence, compliance literacy, who writes the code, how the price is structured, and who owns the IP on day one. This guide give
Most AI features in regulated products die in the gap between the demo and the audit. The model works, the founder shows it to the board, and then someone from compliance asks a question the system ca
The AI prototype looks great in the demo: a notebook, a few example inputs, an impressive output in the board deck. Then months pass and it never ships, or it ships and falls over. This is the most co
Retrieval-augmented generation is the default architecture for most production LLM applications now, and for good reason: it's how you get a model to answer from your data instead of its training set.
Every article on this blog is written by someone who built the thing; not a content marketer reading docs. Real lessons from production, written down while they are still fresh.
Book a call, or get in touch. Both routes end with a straight answer.
Schedule a meeting