Launching Brilliant AI — and why we didn't build another chatbot
Most AI products sold to small businesses are a single widget bolted onto an inbox. We built something more like an operating system.
After two years of building production AI systems in regulated industries, I kept seeing the same pattern. Founders would install a chatbot, get excited for three weeks, then watch it slide back into the corner of the dashboard.
The chatbot was never the problem. The problem was that the business was running on twenty-seven things — a CRM, a calendar, an inbox, a spreadsheet, a WhatsApp group, a notebook in the desk drawer — and a single AI widget plugged into one of them couldn’t move the needle.
So we built something different.
Not a chatbot. An operating system.
Brilliant AI is a coordinated team of specialised agents — receptionist, recall, reviews, finance, marketing, compliance — sharing one customer record, one tone of voice and one set of safety rules.
The receptionist agent doesn’t just answer messages. It checks the calendar before quoting a slot, it flags a returning customer’s preferences, and it hands off to the recall agent when the conversation turns to a missed appointment. The recall agent doesn’t blast templates; it consults the audit log, picks the right channel, drafts in your tone, and waits for your sign-off if it’s not sure.
That coordination is the product. Single agents are easy; an agent team that doesn’t step on each other’s toes is the hard part.
Controlled rails, not off-the-leash AI
The dirty secret of agentic AI right now is that most demos hide the guard-rails. The ones that work in production have:
- A data gate: each agent can read only the fields its role is allowed to read. A customer-facing agent never sees an owner-only note about late payment.
- An action gate: side-effecting writes route through a policy that decides automatic / draft-for-review / block.
- A fabrication guardrail: replies containing a number or claim that can’t be grounded in your business data are blocked, not sent.
- An audit trail: every read, write and decision is recorded with the agent, the input, the disposition and the reason. You can replay it.
None of this is glamorous. All of it is necessary.
For the founder, not the engineer
You don’t write prompts. You don’t manage context windows. You don’t maintain a “RAG pipeline”. You describe your business in plain English; we assemble the team. You stay in charge.
The first week, everything’s drafts. You review, tune the tone, point out what’s wrong. By week two, you trust some channels enough to flip them to auto. By month three the agents know your customers, your suppliers, your seasonal patterns and your preferences.
That compounding personalisation is, in our view, the real moat. A generic AI is a commodity. An AI that’s spent six months learning your business is irreplaceable.
What we’re shipping next
This launch post lands alongside:
- A real-time agent dashboard for the owner
- A WhatsApp channel so customers can text the agents directly
- A native mobile app for SMB owners to see their CRM on the go
If you’re a small UK service business and any of this lands — get in touch. We’re picking the next ten tenants now.