Custom AI Agents

If you can describe the job, we can build the colleague.

Every business has one — the process that only runs because you run it. The January records chase. The Monday report nobody else can compile. The quote that needs three systems open at once. We build agents that do that job the way you do it, on the same engine as our admin agent — which means the same rule applies: it works, you approve.

Yours
Built around your process, not a template with your logo on it
Fixed
Scoped and priced in plain English before we build
Gated
Approval checkpoints wherever work leaves the business
The engine behind it

Our flagship admin agent is built to read inboxes, draft quotes, chase the £480 invoice and wait for a nod before anything goes out. A custom build is that same engine, retrained for your job — not a prototype we're figuring out on your time.

The shape of the work

One job, or a whole relay of them.

  • 01Runs the accountant's January: chases every client's missing records, escalates the stragglers, tracks who's still sitting on their P60
  • 02Compiles the Monday report from the systems it lives in — done before you're at the desk
  • 03Watches supplier prices and flags the changes worth caring about
  • 04Builds the onboarding pack the moment a new client signs — letters, forms, first-week checklist
  • 05Reconciles the spreadsheet against the invoices and lists what doesn't add up
  • 06Multi-agent relays: one agent reads the enquiry, one prices it, one drafts the booking — each handing to the next, the last one handing to you
  • 07Every outward step gated on your approval; every run logged so you can see its working

These are the kinds of builds we scope — not client case studies. When we have named results to show you, we'll name them.

How a build runs

Scoped like a job, not like a project.

01

Scope

You describe the process; we ask the awkward questions. You get a one-page scope: what it does, what it costs, when it's live. Fixed.

02

Build

We build it against the tools the process already uses. Complicated stays our problem.

03

Teach

It learns from your documents, your history, your corrections — and you watch it do dry runs until you'd trust it with the real thing.

04

Live

It takes the job over, approval gates on. Support included at launch, Tune-up after — and as it earns trust, you choose what to hand it next.

Handling sensitive records? Every build can run on our Private AI setups — including one where nothing leaves your building.

Pricing

Sized by the job, fixed at the scope.

Every build is a fixed price, agreed in writing at scoping — the number on the scope is the number on the invoice.

Core

Agent Core

£2,800 one-off

  • ScaleOne agent, one job
  • TriggersRuns when you tell it
  • IntegrationsOne system
  • Approval gatesOn every outward step
  • ReportingRun log
  • Typically live in~3 weeks
  • Launch support1 month
Book a scoping call

Complete

Agent Complete

from £11,000 scoped per build

  • ScaleMultiple agents working a relay
  • TriggersRuns the process end to end
  • IntegrationsWhatever the process touches
  • Approval gatesYou set the gates — and can loosen them as trust grows
  • ReportingFull audit trail of every decision
  • Typically live inScoped per build
  • Launch support3 months
Book a scoping call

The Tune-up

from £400/month

Monthly review of every run, retraining as your business changes, small extensions folded in as you think of them, and a plain-English report of what your agent did with its month. Cancel any month.

Complete builds vary too much for a flat list price — the scope names your number before you commit to anything.

Questions

Asked before, answered straight.

How is this different from your admin agent?

The admin agent is our flagship subscription — a ready-made colleague for inbox, quotes, invoices and diary, one flat monthly. A custom agent is a one-off build for a process the flagship doesn't cover, shaped entirely around how your business runs. The two are designed to sit side by side, so you can run both — but if you're choosing, start with whichever is hurting most.

What can't an agent do?

Judgement calls, and we're not interested in pretending otherwise. An agent is at its best on the repeatable part of a process, and built to stop at the part that needs a human — pricing the weird job, handling the upset client, deciding the exception. That stop is designed in, not a shortcoming we hope you won't notice.

Does the approval rule apply to custom builds too?

Yes, and it's not optional at launch. Anything leaving the business — an email, a quote, an order — waits for your nod. Once it's earned trust on a task, you can loosen the gates task by task. You'll never find out an agent could send things because it already did.

What happens after launch?

A support period is included in every build, then the Tune-up retainer keeps it current as your prices, services and processes change. No retainer? It keeps running as built, and you can pick support back up whenever. Your build doesn't hold you hostage.

Describe the job. We'll tell you if an agent can do it.

Twenty minutes, straight answers — including "you don't need us for that" if you don't.

Zone Ten handles it.