AI your team can check, with your data kept out.
We set up AI assistants for bounded tasks in your team's work, test them on examples before they go live, and train your people to check what they produce. We use AI tools where they are reliable, people decide and check everything, and your data stays out of them unless you agree. This is part of our workflow and automation work, for small businesses, charities and programme teams in London and across the UK.
Who is an AI assistant set-up for?
It suits a team that wants the useful part of AI tools without guessing at the risks. If any of these sound familiar, it is worth a call:
- You already pay for AI tools that hardly anyone on the team uses.
- People are trying AI tools on their own, and you are not sure what data goes into them.
- A task takes hours of reading or sorting each week, such as gathering your sector's news from sources you trust.
- Your board, your funders or your clients have asked how you use AI, and you want a clear answer to give them.
What you get
A short list of the tasks where an assistant helps, and the tasks that stay with a person.
The assistant set up for those tasks, in tools you already pay for wherever they can do the job.
A test on examples you choose, before it goes live. Where it falls short, the step stays with a person.
Automatic checks where they can be built. In our own morning briefing, a script checks every quote and link before anything is published.
The written scope says what data the assistant may see. Nothing goes in without your agreement.
Short training sessions built around your team's own work, and plain-English instructions.
Every file we make is yours, including the written instructions the assistant runs on.
How it runs
A call about the task and the data around it. We ask what the task is, who does it now, what data it touches, and what a good result would look like. We do not talk price on the call.
We ask about your data first. If the answer is no AI on your data, we design the work without it, often with plain automations instead.
A written scope before any work starts: the tasks, what the assistant may see, who checks its work, the dates and the price.
Set up and tested on invented data or on examples you have approved, before it touches live work.
A weekly look at real progress, with the assistant's output side by side with a person's.
Training and handover. The person who checks the assistant's work is trained first, and your team runs it from there.
What to bring
- The task you have in mind, and an example of good work for it.
- The AI tools you already pay for, if any, and who uses them now.
- Your rules on data: what may leave your accounts, and who decides. If you have none written down yet, the call is a good place to start them.
- The person who will check the assistant's work day to day.
An example
This one is real, and it was built by Rron Imami for his own daily reading, not for a client. Overnight it collects new posts from sources he chose. A language model picks the day's items and quotes them, then a script checks every quote and link against what was collected. If a check fails, the day's page is held. What was left out is shown too, with reasons, and a suggested new source joins only on his yes.
It has run since September 2026. For a leadership team or a charity that needs its sector's news each morning from sources it already trusts, the same pattern applies: the model does the reading, and a person reads the result before anyone relies on it, with the checks in between. You can see the morning briefing on the Work page, drawn from its real structure.
AI drafts and your brand
If the assistant drafts anything your audience reads, it should follow your brand's voice. Rron's own workspace checks every draft against a style guide before it goes live, and a person still looks. The same studio designs brand identity, so the voice and the checks can be written together. Charities and programme teams will find more on working with charities, including how consent and funder reporting fit in.
Questions about AI assistants for teams
What can an AI assistant do for a small business?
An AI assistant can take on bounded, repeat tasks for a small business, such as sorting what comes in, picking and quoting items from sources you chose, or writing a first draft that a person then edits. It does not make decisions: people decide and check everything.
Does our data go into the AI tool?
Not unless you agree. We ask first, the written scope says what the assistant may see, and if the answer is no, we design the work without AI.
Who checks what the assistant produces?
A person on your team checks it, every time. We train that person, and where we can, we add automatic checks as well, such as checking every quote against its source.
What happens if the AI gets something wrong?
A person on your team sees it before anyone relies on it, because every output is checked. Each AI step is also tested before it goes live, and where it falls short, the step stays with a person. Automatic checks catch some mistakes before anyone sees them, and the person checking is there for the rest.
Can you train our team on AI tools we already pay for?
Yes, we run short sessions built around your team's own work, so people learn the tool on tasks they really do.
Do we have to use AI at all?
No, a clear process, one database and plain automations often do the job on their own. We would start by mapping the process, and we suggest AI only where it is reliable and you agree.
Book a call
Tell us which task you have in mind. A call with Dorotea and Rron about the work, the data around it and who would check it. Before any work starts, you get a written scope.
Not sure which? Plan your project