One workflow
A useful slice of one real agent—not a vague review of everything.
5 no-cost founding proofs
Let's find out what your AI agent really does.
I'll turn one real concern into a bounded proving sample: five named cases, repeated trials and a plain-English findings note.
FOUNDING OFFER — The first five selected teams receive the Focused Proof, normally £495, at no cost. The other prices show our intended standard pricing after the founding round. No payment details, obligation or paid upselling during your proof.
✓ AI-assisted. Human-accountable.
No payment details.
You speak to the person
running the work.
Proving packages / 01
Focused Proof
Scope agreed before testing · No performance claim
WHAT YOU RECEIVE
Small enough to inspect. Serious enough to be useful.
Five named cases
Normal, edge and failure cases agreed before the trials begin.
Three repeats each
Fifteen recorded trials to see whether behaviour holds or drifts.
A clear evidence note
What happened, what the evidence supports and what I would do next.
A GOOD FIRST FIT
Bring one real question, not a polished pitch.
You do not need a safety team or a finished product. You do need something testable and a genuine concern: an agent sending the wrong thing, using a tool it should not, drifting between runs or failing without a useful trace.
- ✓You have a tool-using agent or working prototype
- ✓There is one behaviour or failure you genuinely worry about
- ✓The workflow can be tested in a sandbox, demo or sanitised setting
- ✓You can explain the intended outcome and prohibited actions
WHAT HAPPENS AFTER
No obligation. The evidence decides the next step.
The sample is enough
You keep the findings note. I tell you honestly if I do not think more testing would add value.
A deeper run is warranted
A paid bounded run starts at £495 for 5 named cases and 15 recorded trials, agreed before work begins.
The problem needs someone else
I say so. This is proving work—not certification, penetration testing or a promise that an agent is safe everywhere.
THE HONEST BOUNDARY
Where failure has a value.
A pass means the agent passed the named cases under the recorded settings during the run. A useful failure means you found a weakness with evidence before relying on the workflow more heavily.
I will not call either result a certification, universal safety claim, security audit or warranty. The conclusion only goes as far as the evidence.
HUMAN ACCOUNTABILITY
Run personally by Jamie—an infrastructure electrician and independent builder of governed automation systems. AI assists the work, but Jamie remains responsible for the scope, trials, evidence and findings.
FOUNDING APPLICATION
What would you like proved?
Tell me about one workflow and what worries you most. I'll review every enquiry personally and reply by email. Five suitable founding proofs will be selected—not simply the first five forms submitted.
No production secrets needed.Start with a plain-English outline. Testing is limited to sandboxed, synthetic or sanitised workflows agreed in advance.
Privacy & testing boundaries
We use the details submitted here only to assess and reply to your enquiry. They are not added to a mailing list or sold. Unselected enquiries are deleted within 90 days. Selected work receives a separate written scope covering authorisation, data handling and retention.
No systems are accessed or tested from this form. Any selected proof begins only after written permission and an agreed scope, using sandboxed, synthetic or sanitised workflows. A proof is evidence about the agreed cases—not certification or a claim that an agent is universally safe.