Hello. We're Knowledge Tree and Butterfly.
Most scientists are right to distrust ‘AI’
Chances are, what your organisation is offering as AI isn't up to scratch. Only 1%1 are using AI effectively to transform their work.
Join the 1%
Special limited-time offer, for sessions booked by 31 December 2026: up to four one-hour sessions at a discounted rate of only £99/session, on one of your own pieces of work. You book and pay for the first session only, and the rest only if you want them. Both buttons open an email to us, copied to Rabbie, our robot, who keeps the diary.
Not yet allowed to use agentic AI at work? See how to land better.
What separates ‘AI’ from AI
| ‘AI’The chatbot your organisation bought | AIAgentic AI, set up for your work | |
|---|---|---|
| What it costs | A licence on every seat, used or not. | A comparable subscription. The real investment is setting it up, once. |
| What it can do | Answers, drafts and summarises what you paste in. | Reads your files, writes and runs code, tests its own output and fixes what fails. |
| How you use it | Ask, copy, paste, repeat. | Hand over a brief and a folder. Come back to a draft. |
| What it remembers | Nothing once the chat ends. You explain your project again every time. | Your notes, conventions and history, kept in the folder. Tomorrow starts where today stopped. |
| Who does the work | You do, and it proofreads. | It drafts, and you review, with every step shown. |
| What you gain | Minutes on emails and summaries. The admin stays. | Weeks of work in days: a 2020 paper rebuilt with ten more years of data in two to three days. |
| Can you check it? | A fluent answer with no workings. | Every number traced to the code that made it. Every step can be rerun. |
| What you keep | A chat history inside someone else's app. | Plain files in your own folders, yours to reuse with any model. |
Either way, what you share goes to a model provider under your organisation's terms. Your AI policy and your data, below.
Already know what you need? See every way to work with us, or choose one and tell us.
Work you can inspect
- Our founder's 2020 fertility paper, rebuilt: 45 countries, about ten more years of data, two to three days from a cold repository to new figures and two draft papers.
- Free statistics course materials, which show how we teach.
- Who we are: the company, and its founder-director.
- Chatbots and agentic AI: the difference this practice rests on.
Your AI policy and your data
The first call asks what you're allowed to use, and the work is designed within that. Training can run entirely on public or synthetic data.
1. What we mean by agentic AI. By agentic AI, the most powerful and transformative form of AI use, we largely mean AI that is:
- a frontier or near-frontier model;
- given permission to work in a largely autonomous capacity on complex, well-defined, multi-stage implementation tasks;
- working through an agentic scaffold, or 'code harness', with wide tool access by default;
- given read-write access to a growing, locally stored knowledge network that you own and maintain.
Each part is explained in plain words in our glossary. This is the AI in the right-hand column above. At present only about 1% of people who use AI use it in this way, most of them in software rather than science. See the waffle chart, with links to the original sources. Back to the text
Or write to hello@ktandb.co.uk, or call or WhatsApp +44 7533 954328 if you'd rather talk first. If we're with a client, leave a message and we'll call you back.