Notes on which AI is worth building, and what happens when it is not.
Four clusters. Every article belongs to exactly one of them, because a pile of unsorted posts is volume rather than authority.
AI opportunity
How to work out which AI projects inside a company are worth doing, and which ones quietly are not.
Building with AI
Shipping AI into production software: agents, automation, evaluation, and what it costs to run once real traffic arrives.
Technical due diligence
Evaluating an early-stage engineering team and codebase, written for the people writing the cheque.
AI training
Teaching a team to use AI on its own work: what it costs, how to choose a provider, and how to tell afterwards whether it changed anything.
Buying AI help
A firm, a freelancer, an agency or your own team. What each really costs over a year, and how to tell a good one from a confident one.
Operator notes
Build logs and the specific decisions behind them, including the ones that went badly.
Everything, newest first
RSSThe AI due diligence question bank: what to put to the founder
Thirty questions for the technical founder, grouped by what each one is really testing, and the follow-up that catches a rehearsed answer.
The AI project brief template: fill this in before you ask for a quote
A one-page template with the nine fields that make three quotes comparable, what to write in each, and the two answers that are allowed to be "we do not know".
What technical due diligence on an AI startup costs
What firms charge for a technical read on an AI company, what moves the number, and the three things a cheap memo leaves out.
The AI vendor scorecard: compare three firms on one page
A scoring template for comparing AI consultancies on the twelve things that predict whether an engagement works, with what each score means.
How long an AI project really takes
Realistic durations for an assessment, a build and a first production release, and the four things that add months without appearing on any plan.
How to brief an AI project so the quotes are comparable
Five things to write down before you contact anyone, and why a vague brief produces three quotes that differ threefold for reasons nobody can explain.
How to get budget approved for an AI project
What a finance director actually needs to see, the three numbers that carry an approval, and why the AI framing is usually what sinks it.
How to tell whether an AI startup's moat is real
Four kinds of defensibility an AI company can claim, which ones survive a competitor with the same model, and the test that separates them in one question.
Our AI pilot failed. What now?
The four reasons pilots die, how to tell which one killed yours, and why the second attempt is usually cheaper and more likely to work than the first.
Questions to ask an AI consultancy before you sign
Eleven questions that separate a firm with judgement from a firm with a deck, and what a good answer to each one sounds like.
Technical due diligence checklist for an AI startup
The twelve questions worth asking before you back an AI company, what a good answer sounds like, and the two that most founders cannot answer at all.
What AI consulting actually costs
Real bands for assessments, builds, retainers and training, what moves each number, and why almost nobody in this market will tell you before a sales call.
What an AI opportunity assessment actually includes
What two to three weeks of assessment buys, what the document contains, and how to tell a real one from a sales exercise with a report attached.
Personal AI agents at work: Muse, Dots, Instinct, Grok Bot
Four agents that act rather than answer, what each can reach inside a company, and why your audit log cannot tell them apart from the person who installed them.
Best AI consultancies for startups, 2026
Seven AI partners a startup can actually hire, what each one is genuinely for, and the two questions none of them answer on their websites.
Best AI training providers for teams, 2026
Six ways to train a team on AI, what each one actually teaches on, and why the productivity figures in this market are so hard to compare.
Is Omarchy the AI-native OS for power users?
Omarchy puts coding agents in the operating system rather than in a browser tab. That is the right direction, and it makes one missing piece obvious.
OpenClaw vs Hermes vs Grok Bot, for companies
The three agent harnesses compared on the questions a company has to answer rather than a hobbyist. Hosting, credentials, real cost, and who fixes it.
What is a 20x company, and how do you become one?
A 20x company does the work of a firm twenty times its size. The gap between the companies that got there and the 89% that did not is not the tooling.
What is an agent harness?
The harness is everything around the model that lets it act on its own. It decides more about whether an agent works in production than the model does.
AI bootcamp, workshop or course?
Three formats, what each one is actually for, and how to pick. Includes the case where the answer is none of them.
AI freelancer or a firm?
A good freelancer is the best value here and the highest variance. How to spot one, and the three jobs where a firm is worth the premium.
How to automate a startup, piece by piece
The order to automate a small company in, why that order is the opposite of what most founders start with, and the test each piece has to pass first.
What corporate AI training costs
Published ranges for AI workshops, bootcamps and executive sessions, plus the six things that move the number and the two lines most quotes leave out.
What an in-house AI team really costs
The twelve-month cost model, including the six lines that never appear in the business case. Put your own salary figure in and the arithmetic is yours.
How to choose an AI consultancy
A four-stage way to evaluate an AI firm, the questions that separate them in twenty minutes, and the five mistakes buyers make most.
How to choose an AI training provider
Nine questions to ask a provider before you sign, what a strong answer sounds like next to a weak one, and the three claims worth ignoring entirely.
How to prove AI training worked
Every provider agrees follow-up matters and almost none of them measure anything. A method that takes one afternoon before the session and one after.
Should a small startup build a software factory?
A software factory multiplies whatever discipline you already have. If you cannot tell whether your developers are doing a good job today, more output is worse.
Who in your company needs AI training first
Four groups, four different sessions, and why putting them in one room serves none of them. Plus the group most companies train last and should train first.
Which processes to automate with AI first
The candidates are the same at every software company. Four tests that predict which survive production, and the order to attempt them in.
How to add AI to your SaaS product
A sequence for adding an AI feature to a product that already has customers, and the four things that decide whether it survives its first month.
How to build an AI MVP
The usual MVP advice applies, plus one question it does not cover. An AI MVP has to validate demand and accuracy, and most founders only test the first one.
How to prioritise AI use cases
A six-step framework for ranking AI candidates, including the step every other framework skips. Where the business value number actually comes from.
How to measure a process nobody has ever measured
You cannot prove an AI project worked without a baseline, and most internal processes have none. Four ways to build one in a fortnight.
How to assess a startup codebase in two days
What technical due diligence can establish in 48 hours, what it cannot, and the six questions that separate a platform from an API wrapper.
How to decide what AI not to build
Most AI opportunity assessments produce a list of things to build. The useful half is the list of things not to build, and here is how to produce one.
What happens when your AI agent gets it wrong
Agents in production are judged on their failure path. What to build around the model so the eleventh real user does not find the edge your ten tests missed.