Agentic AI, hands on.
Most people use a language model as a chatbot. We are going to run one as an agent: a model in a loop, calling tools, doing the actual work. You configure it on your own laptop, in the room, alongside everyone else.
£10 per person. Run at community centres and corporate offices. Date and venue to be confirmed, and you will hear first if you register.
Six things, working, before you leave
Not slides about agents. Each of these ends up running on the laptop you walked in with.
OpenWork, locally
Open source and local first, so what you build does not stop working the month a subscription lapses.
Chrome under agent control
It navigates, fills forms and screenshots what it did, so you can check the work rather than trust it.
Spreadsheets, Drive, GitHub
Connected through MCP servers, including a custom one we write together during the session.
Skills and standing instructions
Context that persists between chats, so you stop re-explaining yourself every morning.
Choosing a model
Kimi K2.6, Qwen, DeepSeek and fifty other providers. What each one is good at, and where each falls over.
Token economics
Where inference spend actually starts, and how to keep it flat instead of watching it climb.
A report that currently eats a morning
We take a piece of recurring work and hand it to the agent. It gathers the source material, fills a spreadsheet with one row per finding and a link back to where each number came from, drafts the summary, and screenshots its evidence as it goes.
Then we watch it get something wrong, which is the more useful half of the session. An agent that cites its sources is one you can check. An agent that does not is one you are simply believing.
Why this example
It touches every piece we configured: the browser, a spreadsheet, persistent context, a custom tool, and a real bill at the end of it.
It is also the shape of most useful automation. Gather, structure, summarise, show your working. Once you have watched the loop run, you can point it at your own version of the same problem.
Bring a task you actually repeat · We will point the agent at it
Run it on weights you choose
The same agentic loop runs on open-weight models you host yourself or call cheaply through a provider. We spend real time on what each option costs per run, because inference spend is where most of these projects quietly die. You leave able to make that call yourself rather than taking anyone's word for it.
Who it is for
Anyone who has used a chatbot and wondered what sits underneath it. Analysts, operators, founders, students, and developers who have not touched agents yet.
£10 per person · No coding background needed · Bring a laptop and a charger
What we will not do
We will not show you how to automate a platform whose terms forbid it. Agents that scrape or mass-message tend to end in a restricted account, and it is not the interesting part of this technology anyway.
Public sources · Your own accounts · Your own data
Hosting a room? We will come to you.
We run these at community centres and corporate offices. If you have the space and the people, tell us and we will bring the session.