AI & Automation
AI agents that do real work, within limits you set
An AI agent is software that reads an incoming email or enquiry, works out what it means, and acts on it — drafts the reply, updates the diary, logs the details — and asks a human before anything consequential happens. Built and supported by the same Bournemouth team that’s looked after local IT since 1995.
- ● Trading since 1995
- ● 4.9-star Google rating
- ● Human approval built in
/01 — THE JOBS THAT QUIETLY EAT THE WEEK
The jobs that quietly eat the week
Every business has them. Enquiries that sit in an inbox until someone has a spare minute. Details typed into one system, then retyped into another. Replies that start the same way every single time. None of it is difficult — it’s just constant, and it lands on whoever happens to be nearest.
An AI agent is a practical way to hand that work over. Not a chatbot bolted onto your website, and not a promise to “transform” anything — a piece of software, built around the systems you already use, that does the repetitive middle steps of a job and brings a human in at the moments that matter. This page explains what that means in plain terms, walks through a worked example, and is honest about the limits. If you’d rather start with the bigger picture, our AI services overview covers everything we build.
/02 — WHAT WE MEAN BY A BOUNDED AGENT
What we mean by a bounded agent
“Agent” has become one of those words that means whatever the person selling it wants it to mean, so here is our definition. An agent is software that can read information — an email, a form, a record — decide what it means using AI, and then act on it: draft a reply, update a record, schedule a message, book a slot.
Bounded means the acting part is deliberately limited. Every agent we build works from a fixed list of allowed actions, in named systems, under rules you set. It cannot invent new abilities, touch anything outside its brief, or send anything consequential without approval. If a job falls outside its list, it stops and hands over to a person — it never improvises.
That constraint isn’t a limitation to apologise for. It’s the reason the whole thing is trustworthy enough to put anywhere near your customers.
/03 — AN AGENT IS NOT A CHATBOT
An agent is not a chatbot
A chatbot answers questions — someone asks about opening hours, it replies, and that’s the whole job. An agent completes tasks. It might never talk to anyone: it can sit behind a shared inbox reading, sorting, drafting and updating records all day without a chat window in sight.
The two can work together — a website assistant that answers questions and then actually books the appointment is a chatbot front with an agent behind it. But you don’t need to pick a technology, and we’d rather you didn’t try. Describe the job in ordinary words — “quotes take too long to go out”, “nobody updates the spreadsheet” — and we’ll tell you which parts suit an agent, which suit ordinary automation, and which should stay with a human.
/04 — A WORKED EXAMPLE, STEP BY STEP
A worked example, step by step
Here is a real example of the kind of workflow we build — including the point where you approve it before anything is sent:
- 1
An enquiry arrives
a web form or email, at 9pm on a Tuesday, long after everyone has gone home.
- 2
The agent reads it
and extracts the useful details: who it’s from, how to reach them, what they’re asking for and how urgent it sounds — even when the message is three rambling paragraphs.
- 3
It drafts a reply
in your tone, answering what can be answered and asking for whatever is missing.
- 4
Nothing is sent.
The draft waits for approval — a one-tap yes, or a quick edit, from a phone over morning coffee.
- 5
On approval
, the reply goes out, the enquiry is logged in your CRM or job sheet with every detail captured, and the right person is notified to follow up.
Notice where AI actually sits in that chain: reading a messy email and drafting a decent reply — the steps that need judgement. Everything else — sending, logging, notifying — is deterministic automation, meaning the same input produces the same result every time. We use AI only for the steps where it genuinely helps and reliable, boring automation for the rest, and we’ll always tell you which is which.
/05 — THE SYSTEMS AGENTS PLUG INTO
The systems agents plug into
Agents are built around what you already run — we don’t start by asking you to change software. Typical connections include:
- Email and shared inboxes — reading, sorting, drafting
- Website forms and enquiry pages
- Calendars and booking systems
- CRMs, job sheets and customer records
- Spreadsheets and simple databases
- Messaging — Slack, Teams and SMS — for notifications and approvals
Where a system has a proper integration, we use it. Where it doesn’t, there is usually a sensible route in — a monitored inbox, an export, a form. Occasionally the honest answer is that a system can’t be connected safely, in which case we say so before you spend anything. And these aren’t hypothetical connections: booking automation, review requests, SMS scheduling and Slack-connected workflows all run in our own business every day, built and maintained in-house.
/06 — LIMITS, APPROVALS AND THE PAPER TRAIL
Limits, approvals and the paper trail
Human control isn’t a feature we bolt on if you ask for it — it’s how every agent is designed from the first conversation. Each one comes with:
An allowed-actions list
the complete set of things it can do, agreed with you before anything is built.
Approval gates
on consequential steps — anything sent to a customer, and anything that commits money or time, waits for a human yes.
A log
of what it did, when, and what it was acting on, so its work can be checked.
A pause switch
the whole agent can be stopped at any time without breaking anything else.
You choose where the gates sit. Some businesses keep approval on every outgoing message; others let routine steps run on their own and review only the exceptions. Either is fine, and it can be adjusted as your confidence grows — that’s a conversation, not a rebuild.
/07 — PROOF YOU CAN LOOK AT TODAY
Proof you can look at today
We’ll be straight with you: AI agents are new enough that we don’t yet have years of customer case studies to point at, and we won’t invent any. What we can show you is working software.
The 365 AI OS is our Claude-powered AI desktop prototype — a live, honestly-labelled scripted demonstration of an agent carrying out multi-step work, which you can watch in your browser right now. Our custom monitoring dashboards and the 365 PC Manager app are real products we built and run, and the automations described above — bookings, review requests, SMS, Slack workflows — handle our own daily operations. We use this technology on ourselves first.
Behind it is a family-run firm trading since 1995 with a 4.9-star Google rating, a Microsoft Partner, recognised as a Victron Recommended Software Integrator for our telemetry integration work. If your agent misbehaves at 9am on a Monday, you phone the same local people who look after your IT — not a ticket queue in another timezone.
Open the 365 AI OS demo/08 — WHAT SHAPES THE QUOTE
What shapes the quote
Every agent is priced the same way: a monthly service subscription that keeps it running, monitored and supported, plus a one-off design and build quoted by complexity. There are no tiers to decode, and no figure on this page we’d quietly change later — the build price depends on the job, so we confirm it with your quote.
What moves the number:
- How many systems the agent connects to, and whether ready-made integrations exist
- How many distinct actions it performs
- Where the approval gates sit, and how many people are involved in them
- Volume — ten enquiries a week is a different build from two hundred a day
- How sensitive the data is, and what handling that demands
The first conversation doesn’t require you to know any of the vocabulary on this page. Tell us what’s wasting your team’s time and we’ll come back with what an agent could take off it — and, just as honestly, what it shouldn’t.
// SERVICE & BUILD
Monthly service — confirmed with your quote
One-off design and build — quoted by complexity Every agent runs as a supported monthly service. The design and build is a separate one-off cost, quoted by complexity once we understand the job — no tiers, and no figure on this page we would quietly change later.
// GOOD QUESTIONS
Frequently asked
Will an AI agent email my customers without anyone checking?
Not unless you decide it should. By default, anything customer-facing waits for human approval — the agent drafts, a person taps yes or edits first. Some businesses later let routine, low-stakes messages go out automatically and review only the exceptions, but that is your call to loosen, never our default.
What happens when the agent gets something wrong?
Because agents are bounded, a mistake is contained: the worst a drafting error can do is produce a bad draft, which the approval step catches. Every action is logged, so we can see exactly what happened and adjust the agent’s instructions. And because it’s supported by the same local team that looks after your IT, reporting a problem is a phone call, not a ticket into the void.
Do we have to change the software we already use?
No — building around your systems is the whole point. Agents connect to the inbox, calendar, CRM and spreadsheets you already run. If a tool genuinely can’t be connected safely, we’ll tell you before any work is quoted, not after.
How is this different from just using ChatGPT?
A chat window can draft text, but you do all the carrying — copying details in, copying answers out, updating your systems yourself, every time. An agent is wired into those systems, follows your agreed process the same way on the hundredth run as the first, and takes actions within set limits, with approvals where they matter. A chat tool is something a person must operate; an agent quietly runs the job.
Is our business too small for an AI agent?
Size isn’t the test — repetition is. If a task follows a similar pattern many times a week and eats someone’s time, it may be worth automating whether you employ two people or forty. We’re a family-run small business ourselves, and the automations we run daily were built for a team our size. We’ll also tell you honestly when a job is too small or too irregular to justify a build.
What about our customer data?
Data handling is agreed as part of scoping, not bolted on afterwards. We define exactly what the agent is allowed to read, keep information in your existing systems wherever possible, and avoid moving anything it doesn’t need. Data protection obligations are discussed openly at the quote stage, so you know precisely what goes where before anything is built.
See what a bounded agent could do for you
Tell us about the process, the systems and where a person must stay in charge — we’ll design the agent around that.
01202 775566 · help@365techies.co.uk · MON–FRI 9AM–5PM