When you build a team of AI agents, how do you work out who should be on the team, and how do you give each of them the skills and knowledge they need?
That's the job of Aero, our auto-configurator AI, when it sets up a new Airgentic service. Here's how it goes about it.
1. It starts with the business, not the website
Before choosing a single agent, Aero works out what the organisation most needs its website to achieve. It reads the homepage calls to action, the about and strategy pages, and looks at where the content sits. Is the priority donations? Fewer calls to the council? Getting buyers to a dealer? Those imperatives decide everything that follows.
2. It lists the jobs visitors actually come to do
For each imperative, Aero lists the real visitor tasks behind it, written in the visitors' own words. It also looks for tools the site already offers, like a bin-day address search or a dealer locator. Each of those usually has a data source an agent can use.
3. Every agent has to earn its place
Aero tests each candidate agent against one question: what can it do that a general assistant can't? The answer might be:
- a live lookup
- deep knowledge of one process
- getting a motivated visitor to the right form
- a sensitive topic that needs a gentler tone
- keeping a flood of job ads out of everyone else's answers
If an agent can't answer that question, it doesn't make the team.
4. Each agent gets the knowledge of a skilled staff member
A rates agent doesn't just search the website. It knows how a rates notice is put together, every way to pay, the concession and hardship options, and what the council doesn't decide. Figures and dates are always looked up and cited, never recalled from memory.
5. Each agent knows where its job ends
Clear boundaries keep each question with the right specialist. A gift made today in someone's memory goes to one agent; a gift left in a Will goes to another.
And every agent finishes at an outcome the organisation cares about: the right donation form, a dealer nearby, or an enquiry passed to the right team, with the visitor's permission.
6. The team defines the tests
Each agent's best question becomes a welcome example. Every boundary between agents becomes a routing test, and every lookup is tested both with real inputs and with inputs that should find nothing.
The result
A team of five to eight specialists, each built around what the organisation needs most.
If you'd like to find out more, get in touch.
