How to estimate an AI app project
Why AI features break normal software estimates, which parts are genuinely predictable, and how to put a defensible range on a project before you commit.
Rohan SahuLong read
Read the article Engineering notes
Insights from what we actually build
Long-form notes on the parts of AI and software delivery that are hard to get right — written by Rohan Sahu from systems this team built, runs and has broken.
We publish an article only when we have first-hand experience of the thing it describes. That keeps this list short on purpose. There is a great deal of general AI advice on the internet already, and adding a summary of it would not help anyone — so what you will find here is limited to work we have done ourselves, including the parts that went wrong.
Production checklist for an AI voice agent
The failure modes that only appear once a voice agent takes real calls — latency budget, barge-in, consent, fallback, monitoring — and what to have in place before launch.
Rohan SahuLong read
Read the article Have a project these questions apply to?
Start with an indicative estimate, talk to our consultation agent, or write to the team. All three go to the same place.