Blog
Write-ups on the parts of AI, infrastructure and automation work that only show up after launch. No trend pieces, no predictions.
An automation is only useful if nobody has to watch it. The ones that end up watched — checked every morning, re-run by hand, quietly distrusted — almost always skipped the same few habits.
Read the postThe managed-cloud bill is easy to resent and easy to compare against a rented server. The comparison that matters is harder: what the managed services were quietly doing for you, and who does it once they are gone.
Read the postInvestors are good at reading financials and markets. The codebase is usually a black box until something goes wrong after the round closes. Ten questions that open the box without needing to read the code yourself.
Read the postMost retrieval-augmented generation projects are tuned by reading answers and deciding they feel better. That works until the second person on the team disagrees. A small, honest evaluation set is the instrument that ends the argument.
Read the postA demo that works is not a system that works. The gap between the two is filled with problems that have nothing to do with model architecture, and teams keep discovering them one sprint at a time.
Read the post