AI workflow design and governance for UK firms ready to move past chat as a shortcut.
From ad hoc prompts to designed systems.
Mapped, governed, shipped.
A large-scale document review pipeline, shipped, live as SaaS.
Prompt engineering curriculum commissioned by KFUPM, Saudi Arabia.
Commissioned engagement at the University of Bristol.
Right Prompt, Right Task is Matthew Thistle’s AI consultancy, installing agentic workflows, custom skills, and knowledge graphs that compress days of work into hours for UK and MENA firms.
Two ways in.
AI Diagnostic Audit
Scoped to the detail you need. For firms of a dozen staff or more.
Three to five stakeholder interviews. A working map of where AI actually fits in your operations, and where it does not. A written deliverable your leadership can act on, with no lock-in to a training package or build engagement.
For most firms, this is the first time AI strategy gets written down.
AI Business MOT
A single-session diagnostic.
One conversation, a clear view of three automation priorities, a written report you can hand to your team or use to scope further work. Suitable for individuals, small firms, or as a feeder into the Audit.
| AI Business MOT | AI Diagnostic Audit | |
|---|---|---|
| Format | One 45 to 50 minute session | Three to five stakeholder interviews over two weeks |
| For | Owner-led firms, sole practitioners, individuals | Firms of twelve-plus staff |
| Deliverable | Three prioritised AI opportunities, written report | Operations map, ranked AI opportunities, implementation roadmap |
| Investment | £97 | Scoped on call |
| Time to delivery | Same day | Two to three weeks |
The methodology underneath every engagement
AIMED is the framework I use for every engagement, from a single-session MOT to an enterprise training programme. It is designed to move a team from discovering AI to governing it at scale.
- A
Actor
Who the AI is being asked to be. Role, perspective, expertise.
- I
Instruction
What you are actually asking for. The task, stated precisely.
- M
Method
How the work should be done. Steps, reasoning, constraints.
- E
End-goal
What success looks like. Provide an example if necessary.
- D
Distil
What to strip out. Length, format, irrelevant detail.
What has been built.
A complex, large-scale document review pipeline. Not just batch processing.
Criteria-driven review, pass or fail determination, specific feedback against named standards. A 15,567% efficiency increase over manual review, with 96% agreement with expert human reviewers at 250+ documents per hour.
The work it replaces used to take a human reviewer anywhere from ten to forty minutes of focused cognitive work per document. That time is now returned to the organisation, every day, at full quality. The architecture transfers directly to compliance checks, quality assurance, regulatory review, and any workflow where criteria-driven judgement meets document volume.
Three production websites built from scratch, Python cloud-bucket architecture, including this one and the pipeline above. An active R&D line building how AI agents can connect to fintech and blockchain infrastructure, with compliant wallet access for autonomous market operations. In development.
See the pipeline in a scoping conversation.
Scope a Diagnostic AuditWhat I run, every day.
I run a coordinated stack of AI agents inside my own consultancy. This is the infrastructure I help clients build, at the scale that fits them.
- i
Research and document agent
A long-context agent handling heavy document work and background research, with a context window large enough to hold whole books.
- ii
OpenClaw mobile production agent
Runs on my own VPS with a full memory hierarchy and a substantial tool suite. Serves as my mobile CRM when I am out with clients, captures and preserves voice notes while I am walking or exercising, and keeps a working memory of every live engagement. A second brain, portable, always on.
- iii
Production automation pipeline
Sits on the same VPS. Orchestrates multi-step workflows across client tools, triggered by email, calendar, document events, or direct instruction. The engine behind the agents above.
- iv
Knowledge base agent
Maintains my wiki as I work, so nothing is lost and everything is retrievable. Self-healing across sessions: recovers its own context from commit history and its own logs, with no handoff required. Holds the deepest context of any agent in the stack.
- v
Specialist document review agent
Builds and coordinates with the review pipeline above.
- vi
Collaborative strategy agent
The partner I use for writing, strategy, and live client work.
All of them coordinate through the same architecture: MCP servers, OAuth into Google Workspace, a Notion database that acts as shared scratchpad between agents, and audit trails on every action. This is not a demo. It is how I operate as one person, every day.
Selected recordings.
The measured efficiency gain on a production document review pipeline. Tailored compliance checking at document scale, criteria-driven review against named standards, 96% agreement with expert human reviewers.
See the case studiesLatest Insights
AI adoption strategy & original thinking
Machines of Infinite Patience: How AI Restores the Sacred Contract of Teaching
A Senior Engineer Would've Found This Bug in Hours. I Found It in 10 Min by Orchestrating 2 AI Agents.
What Happens When Your AI Agent Escapes Its Sandbox?
There Is Something Strange About Sitting Across from a Business Owner and Watching Them Go Quiet
I Just Crossed 1.5 Million Words Dictated by Voice
Something Happened at the End of a 21-Hour Coding Session with Claude Code
The Night I Stopped Writing State Objects by Hand
A Maserati That Could Not Move
If you are ready to draw the map, book the Audit. If you want a single-session view of your situation first, book the MOT.