We are named after the two jobs, because both of them matter.
Hardy & Butler designs, builds and runs AI agents for British businesses. Butler is the working half. Hardy is the half that makes it safe and secure. We put both names above the door, because we include the working agent and the oversight in the same engagement.
The technology stopped being the hard part some time ago.
An agent that can read an inbox, understand it and draft a decent reply is no longer remarkable. That problem is solved, and it is solved well enough that the demo always goes fine.
What is not solved is everything around it. Who signed this off. What did it do at three in the morning. What does it cost to run in March. What happens the day it sends something wrong to your best customer, and who finds out first, you or them. Ask those four questions in a demo and watch the room.
We started Hardy & Butler because those four questions are the entire job, and because a small British company that answers them properly is worth more to a business your size than a large American one that answers the first question beautifully and the rest with a badge.
- 03:12:09 Butler Read the aged debtor list, 41 accounts past 30 days Accounts system, read only. Nobody contacted at this step.
- 03:12:22 Butler Dropped 6 accounts that paid yesterday A reminder sent to somebody who has already paid is the fault that costs you the customer.
- 03:12:23 Hardy Checked 35 drafts against the fence 34 inside the agreed list. 1 held.
- 03:12:25 Hardy Held 1 draft for a person Reason recorded: the account is in dispute, raised 11 days ago.
- 07:40:12 A person Approved 34, amended 1, sent Named in the ledger. The amendment is kept beside the original wording.
An example of the format, not a customer’s data. This is the answer to the second question, and it is the one most suppliers cannot show you.
One person is accountable for every build.
Hardy & Butler is run by James Slater. Seventeen years in the Army, then thirty years running operations and fleet businesses, the last of them as an operations director answerable for processes that had to work on a Sunday as well as a Tuesday. He builds this software for his own companies first, and nothing is sold here that is not already running somewhere it matters.
If you engage us, he is the person who scopes the work, agrees the number that has to move, and answers the phone when it does not.
Four things we believe.
An agent you cannot audit is a liability, not an asset.
If nobody can reconstruct what it did and why, it is not automation, it is exposure. Every agent we build writes down what it did, what it used and who approved it, from day one, not as a later phase.
The human keeps the decision.
Software should do the work and the person should keep the judgement. Anything that leaves your business, spends your money or changes who can see what waits for a named person. That is the default, and you have to ask us to loosen it.
Pay for hours saved, not for revenue we cannot prove.
The honest research says AI reliably saves time and rarely, so far, adds revenue. So we price against hours and cost, we agree the one number that has to move before we start, and we report against it whether it flatters us or not.
Small, live and boring beats large and theoretical.
One agent doing one job in production this month teaches you more than a six month programme. Pilots that never touch real work are one reason Gartner expects over forty per cent of these projects to be cancelled by the end of 2027. We build small, put it live, and then go again.
Audit, build, prove, run.
Audit
Two weeks, fixed fee, credited back. We map the work as it is actually done, rank it by what it costs you, and produce a fixed price.
Build
Three to four weeks for a first agent. Fenced before it is built: what it may do, what needs approval, what it must never touch.
Prove
Hardy measures against the one number we agreed. You get the evidence quarterly, including the quarters it does not move.
Run
Monitoring, tuning and a brake you control. Published monthly price, so it goes in the budget before you commit.
Four things we will not do, whatever the invoice says.
- We will not build an agent that can spend your money on its own authority. Not with an approval workflow bolted on afterwards. Not at all.
- We will not build what we cannot measure. If we cannot agree beforehand on the number that has to move, we have not understood the problem yet.
- We will not sell you a pilot we do not believe will reach production. If it is a science project, we will say so and you can decide.
- We will not lock you in. You own the agent, the prompts, the configuration and the code. If you want to take it in-house or give it to somebody else, that is a handover, not a negotiation.
We will not build you an agent we cannot audit.
On every engagement.
- Human approval by default on anything that leaves the business.
- On every agent we build and run for you: your data is never used to train a model, ours or anybody else's.
- UK data residency for the agent's own records, and a data processing agreement on every engagement. How this website handles an enquiry is set out in the privacy notice.
- Working to UK GDPR.
- A written oversight policy and an audit trail for every agent we deploy.
- You own the agent, the prompts, the configuration and the code. No lock-in.
- A written oversight policy for every agent, naming who approves what.
- A brake you control, tested with you on go-live day.
- Quarterly evidence, including the numbers that did not move.
Tell us about those tasks that never land on time.
You do not need to know what an agent is, how it works, or which one you need. Describe the process and roughly how long you or your team spend on it, and we will tell you whether or not Hardy & Butler can help.