Knowledge Tree
and Butterfly Analysis that shows its working
Explainers

Chatbots talk. Do-bots work.

If AI so far has felt like something done to you — a chatbox bolted onto your tools, an assistant button you didn't ask for — you have met the chatbot. Not the do-bot.

A do-bot is the same model wired into the tools developers already use: files, a shell, tests, version control, permissions. Developers call that wiring a code harness, or a scaffold; when I say agentic AI, this is what I mean. The chatbot answers. The do-bot does the work, and shows you what it did.

Why are do-bots revolutionary, when chatbots (after the initial hype died down) aren't?

Because do-bots can read and understand very complicated briefs, can get familiarised with massive amounts of information, and can generate and course-correct on solutions incredibly fast.

If a chatbot can help you get 10% better at your job, do-bots can help you get 10 times better! (And even 10X is lowballing...)

Of everyone who has used generative AI, only a small fraction has used it this way. Almost all of them write software for a living.

Each dot is a share of people who have used generative AI. A grid of 500 dots, 25 across and 20 down, filled row by row. 464 grey dots (93%) are people who have only used a free chatbot; 31 darker dots (6.3%) pay for a chatbot; 5 red dots at the bottom right (0.9%) have used a model inside a coding harness, a do-bot. Estimates worldwide, March 2026: about 965 million people have used generative AI, 70 million pay, and 9 million (between 7 and 12, so 4 to 6 dots) have used a coding tool. Sources and method on blog.jonminton.net. Each dot is a share of people who have used generative AI. chatbot, free 93% chatbot, paid 6.3% do-bot 0.9%
Worldwide estimates, March 2026. In the UK a larger share has tried a chatbot; there is no evidence the do-bot share is any bigger. Rebuilt from a viral chart that got its numbers wrong — the correction and the sources are on my blog.

If you work with data and evidence, you are almost certainly a grey dot. That is not behind. It is early.

If you build software, you are one of the red dots. The grey rows are where the demand is coming from.

What a do-bot needs from your work to do it well is the subject of the knowledge tree.

Back to the two doors →