The best AI use cases start from a specific task you can check, not from a solution. That was the message of “Don’t Build Another AI Agent”, a hands-on workshop run by Harry Hobbs of Baltic Augment at aiblLIVE London 2026.
Attendees set up an AI interviewer in ChatGPT to question them about their own work. They checked its diagnosis, then turned the learning brief it wrote into an audio explainer in NotebookLM. This guide sets out the method so you can run it yourself.
Choose a specific task from your own week that you would know how to judge. Hobbs asked the room for work that is gruelling or harder than it should be, described without a solution in mind. “Improve recruitment” is too broad. Turning workshop feedback into summary results and charts is the right size.
The bar to build is disappearing, but the bar to production has never moved.
Harry Hobbs, Head of Business Intelligence, Baltic Augment
Sort the task into tool AI or decision AI. Tool AI is lower-consequence work for a bounded task, where you check and own the result, so you can improve it yourself. Decision AI could affect pay, compliance or safety. Baltic Augment leaves that work to the central data team and AI engineers.
Save interviewer instructions in a ChatGPT Project, then type “start my interview” and answer for 10 to 12 minutes. The AI assesses the work pattern, the consequence and the main opportunity. Check its diagnosis against what you said before accepting the detailed learning brief it writes.
Listen against four questions. Does it reflect the task you described? Does it teach you something, rather than repeat you? Does it keep the human verification step? Does it add anything your source does not support? Using your brief as the only source in NotebookLM, with web sources off, cuts that risk.
Human involvement cannot become the person who clicks the button at the end.
Harry Hobbs, Head of Business Intelligence, Baltic Augment
Harry Hobbs is Head of Business Intelligence at Baltic Apprenticeships and consultancy lead for Baltic Augment, which helps organisations through data, AI and automation. With Lead Analyst Faith Khan and AI Engineer Jamie Blake, he ran this workshop on the Workforce track at aiblLIVE London on 20 October 2026.
If your teams are keen to build agents, this is the step to run first. The full guide gives you the method and the prompts to run it.
For managers, HR and people leaders, and anyone commissioning an AI agent. Seven pages, about a nine-minute read.
