Digital HQ

We've Got AI Agents in Production

By Vil Pietersen3 min read

Last night I connected Norwich Executive Travel to Secret HQ, our not-very-secret AI side of Digital HQ. As of this morning, an AI agent is doing real work for a real business.

That's a big deal for us. So I'm writing a post about it.

What it does

Julian runs Norwich Executive Travel. A lot of his bookings still arrive by email: a customer writes in with a date, a time, a pick-up address and an airport, and someone has to turn that into a booking in the system.

That someone was Julian. Every job meant reading the email, finding the customer, typing in the addresses, getting a quote and saving it. He did it day after day, and he hated it.

Now he copies the email, pastes it into Secret HQ, and the agent does the rest:

Pasted email“Stansted, Fri 6am…”LLMreads the email,fills the bookingfind_customer · get_quoteDraftwaiting for JulianCustomerMrs PatelPick-upNR2 · Fri 06:15Drop-offStansted T1Quote£148.00Approved?Booking livedriver assignedThe agent can only create drafts.A person still presses the button.
Paste an email in, get a draft booking out. Julian still makes the final call.

The agent reads the email and works out who the customer is, where they're going and when. Then it uses the same tools the booking system already has, like finding the customer and getting a quote, to fill in a booking. It saves the booking as a draft.

Julian stays in charge

The draft is the important part.

The agent can't confirm a booking, send a driver or charge anybody. It can only prepare the booking. Julian looks at the draft, checks it, and approves it himself. If the agent misread something, he fixes it before it matters.

This is the guardrails idea from our post on how agents work, used on a real job. The agent does the slow part and the person makes the decision. Julian keeps control of his business and stops doing the job he hated.

Why paste first?

Pasting is a stepping stone. The real goal is for the agent to read the bookings inbox itself, so the draft is already waiting when Julian sits down with his coffee.

NowEmail arrivesJulian copies& pastes it inAgentmakes the draftJulian approvesNextEmail arrivesthe agent reads the inbox itselfLLMmakes the draftJulian approves
Now: Julian pastes the email. Next: the agent reads the inbox itself.

Connecting to an inbox properly takes time to get right: permissions, deciding which emails count as bookings, and dealing with replies and forwards. We didn't want Julian to keep typing in jobs while we built that. So we shipped the paste version first. It uses the same agent and the same draft-and-approve step. Only the way the email gets in is different.

Julian gets most of the benefit today. When the inbox connection is ready, it will replace the paste step and nothing else about the process changes.

We've only scratched the surface

This is the part that excites me most. Nothing in the booking agent is specific to bookings. It's the same loop as any other agent: read something, use some tools, prepare a result, and let a person approve it. To give it a new job, you mostly give it a new set of tools.

LLMsame loop,different toolsPasted email → draftlive nowInbox → draftnextQuote follow-upsChasing invoicesCustomer repliesWeekly reports
One agent loop, many jobs. Each new job is mostly a new set of tools.

A few jobs we're already thinking about:

  • Quote follow-ups. Someone saved a quote last week and never booked. The agent drafts a friendly nudge.
  • Chasing invoices. It spots what's overdue and drafts the reminder, so nobody has to have that awkward conversation from scratch.
  • Customer replies. "What time is my driver coming?" The agent can look that up and write the reply.
  • Weekly reports. Jobs done, quotes won, busy days coming up, written in plain English instead of a spreadsheet.

Every one of these works the same way as the booking agent. The agent prepares the work and a person approves it.

Small, but significant

One agent turning emails into bookings for one business won't change the world. But it's real work for a real customer, and he's already happier for it. Until last night, agents were something we wrote about and built demos of. Now one is in production.

Next up is the inbox. After that, we'll see where it takes us.

If you've got a job you hate doing that starts with "someone has to read this and type it into that", get in touch. That's what we're building agents for.

About the author

Photo of Vil Pietersen

Vil Pietersen

Founder, Digital HQ

Vil started out doing high-risk search on operations with the British Army in 2008. From there he moved into web development and SEO, building websites since 2010 and React web apps since 2019, before going to university for a first-class BSc (Hons) in Artificial Intelligence. He now runs Digital HQ in Bristol, building custom software and AI agents you can actually see into.

Get in touch

11 min read

What Is an AI Agent? A Simple Guide If You Already Use ChatGPT

What is an AI agent? If you know how ChatGPT works, you're most of the way there. We build one up step by step: memory, tools, sandboxes and guardrails.

6 min read

AI and Robotics Companies in Bristol: The 2026 Showcase

From AI chips to bionic arms and wind-turbine drones: the AI companies in Bristol and robotics firms putting the city on the map, with links to each.