Agent Building Workshop by Ollo
Sign in to watch the recording
The write-up below stays open to everyone. The video itself needs a free account.
About
AI agents can sound complicated, but most SMEs should start simple: e.g. an agent that answers customer questions, qualifies enquiries, and guides people toward a booking.
Recap
Husayn Kassai, founder and CEO of ollo, led a hands on workshop on building your first AI agent. No slides marathon, just five minutes of context and then everyone built an agent together on a live screen share.
His starting point: a large language model on its own gives you the internet's best average answer. It does not know your business, which customers pay late, or who signs off spend over £5,000. What makes it useful is the loop you build around it, called a harness. The harness controls what the model can see, which tools it can touch, what it may change, when it must ask a person, and how you verify the result. Think of it as hiring someone with 20 years of industry experience. You would not let them loose on day one. You would tell them where to look and when to check with you.
Everyone then built a simple agent live: first one that could only answer "red" or "green" (and answered at random, because that was all it was told), then the same agent with one extra instruction, classify the sentiment of the message first. Suddenly an angry customer email came back red and a happy one came back green. The lesson: an agent is only ever as good as the instructions you give it.
The second half covered a shared invoice tracking agent, and team spaces, which work like rooms in a house. Point the model at the kitchen and it will find the spoon. Give an agent a space full of sample invoices and it will find the unpaid ones, with the supporting email text quoted for each row.
Key takeaways
- Start from a business problem, not the technology. Answering customer questions or chasing unpaid invoices beats a grand automation plan.
- Buy a vertical tool where a strong one exists (Husayn's examples: Jessie for restaurant phone orders, Jelly for trades and home services over WhatsApp). Build your own where your business is genuinely different.
- Around 95 per cent of everyday tasks do not need the most expensive frontier models. Watch your usage, both for cost and for energy.
- Only about 5 per cent of an organisation needs to build agents. The other 95 per cent are users, and they should be trained to read results and understand cost.