40%+
of agentic AI projects will be cancelled by the end of 2027, Gartner predicts, through rising costs, unclear value or weak risk controls.
Source: Gartner press release, 25 June 2025For businesses of 50 to 500 people
Start with one task. Keep your people in charge.
Most AI programmes stall after the pilot. We start smaller: one person, one repetitive task, one agent that drafts while your people brief it and sign it off.
0 min
A free first conversation
£750
The first paid step
£0
New software to buy
Why small first
Big AI programmes stall. Small ones compound.
Pilots everywhere, P&L nowhere. The programmes that fail try to change everything at once. The ones that stick start with one person's week.
+11 pts
Daily AI users are more confident in their job security than infrequent users: 68% against 57%. Use builds confidence, not fear.
Source: PwC Hopes and Fears 2026, 49,364 workers in 48 countries51%
of workers say they have the learning resources they need, down from 59%. Teams want help beside them, not a report.
Source: PwC Hopes and Fears 2026The method
The Delegation Loop. Delegate, don't abdicate.
Your people already know how to delegate. An agent is a capable, tireless junior: it does the drafting, and your person stays accountable for the result. Three of the four steps belong to a human.
Brief
Sets out the task, the sources and what good looks like, the way they would brief a new colleague.
Draft
Does the repetitive part: gathers, sorts, summarises, drafts. Minutes, not hours.
Check
Reviews the output against the brief, corrects it, and sharpens the brief for next time.
Sign off
Approves and sends. Nothing leaves without a named person's sign-off, and every run is recorded.
Which task first
The work that eats a Tuesday.
The best first task is repetitive, written down, and checked by someone who already knows what good looks like. A few typical ones, by function.
- Month-end variance commentaryFrom the trial balance to a first draft for the FD
- Supplier statement reconciliations
- Credit-control chasers
- Proposal and tender first draftsFrom the call notes and past wins
- CRM notes after every call
- Account briefs before a meeting
- The weekly status reportPulled from the tracker, written for the board
- Supplier and order chasers
- Procedure updates after a change
- Job descriptions and advertsFrom the hiring manager's brief
- Draft answers to policy questions, from the handbook
- Onboarding packs
The journey
Start free. Pay as the value shows.
Fixed scope. Fixed price. Every step ends with evidence, measured against your own baseline, and you take the next one only when the last has paid its way.
00 · 30 mins
Clarity Call
We find the one task in your week most worth handing to an agent.
FreeYou leave withYour first task, named
01 · 90 mins
First Agent Session
One person, one task, one working agent, in tools you already pay for.
£750You leave withA live agent and the hours it saves
02 · 3 days
Team Starter
Three days beside your team: 1-2-1s at desks, a team session, first agents for each person.
£5,000 · creditedYou leave withA team using agents and a backlog of next tasks
03 · 2–3 weeks
AI Diagnostic
Workflow mapping, opportunity and risk, an ROI model and a 100-day roadmap.
£5–10kYou leave withA board-ready plan
04 · 30 days
Agentic Workflow Sprint
Production-ready workflows across a function, with a working prototype in week one.
£20–35kYou leave withWorkflows running in production
05 · Ongoing
Fractional AI Operating Partner
We run, govern and grow your agents, and train your own champions to take over.
£8–12k / monthYou leave withCapability your business owns
Team Starter fees credit in full towards a Diagnostic or Sprint booked within 60 days · Prices exclude VAT
Not sure which rung fits? Thirty minutes will tell you.
Book a free clarity call → Explore the services →
Senior time · by the day or the session
Not a project yet? A day on a decision for the business, or coaching for you.
Specialist Days from £1,750 · Executive AI Coaching from £750
Private equity · at the deal or in the portfolio
For private equity →
Start in the data room on a live deal, or with one task in a company you own.
AI Due Diligence · £8–15k · 2 to 4 weeksBeside your people
Keep the judgement. Lose the drudge.
We work in your office, 1-2-1 and with whole teams, until your people build and run their own agents. Then we step back.
- No one is replaced by a tool. The person who briefs and signs off is the expert. The agent can do neither.
- Learning by doing. Each person builds an agent for their own work, not a demo.
- A fixed term, then champions. We train people inside your business to carry it on, so we never become a permanent cost.
One person's week
Before40 hours
After the first agents40 hours
Repetitive workJudgement and relationshipsEverything else
Illustrative. We measure your real baseline in the First Agent Session.
Safe by design
Every agent has an owner, a review step and a record.
Unmanaged personal AI becomes shadow AI. Governance starts in the first session, not after the first incident.
A named person.
Someone is accountable for each agent and for what it produces.
Nothing leaves unchecked.
A person reads, corrects and signs off before anything goes out.
A register from day one.
An agent register, an approved tools list and a log of every run.
| Agent | Owner | Review before it leaves | Runs logged |
|---|---|---|---|
| Month-end variance commentary | Finance director | Read, corrected and signed before the board pack | 14 |
| Proposal first drafts | Sales director | Every draft read before a client sees it | 31 |
Who you work with
Still building agents every week.
Danny Reeves has been CEO, COO and CIO across data centres, hyperscale infrastructure and mid-market businesses. Praxis runs its own operation on the method it sells.
The Sunday editor.
Every week Danny publishes two LinkedIn posts, each with a short video. Reading the week's AI news, choosing the two stories worth a mid-market leader's attention, drafting the posts and producing the videos was a weekend's work, done by hand.
- Brief · DannyA written spec: the reader, the voice, what never goes in, how a story scores
- Draft · early SundayReads the week, picks two stories, drafts both posts, renders both videos
- Check · over breakfastDanny checks the facts it has flagged and edits what needs editing
- Sign off · Tue and ThuDanny posts. The agent publishes nothing
The brief has been rewritten more times than the posts have. Three of the four steps are still his.
Read the posts on LinkedIn →Before you book
Fair questions.
If yours is not here, ask it on the call.
- Do we need new software?
- No. We build in the tools you already pay for, and we say plainly if a task would need anything more.
- Is our data safe?
- Every agent has a named owner, runs on an approved tool and has a review step before anything leaves. The register starts in the first session.
- Will this replace my people?
- No. Your person briefs the agent, checks its work and signs it off. The agent drafts. The aim is less drudge, not fewer people.
- What if it does not work for us?
- Then you have spent thirty minutes, or £750 at most, and you know. You take each next step only when the last one has paid its way.
- How is this different from buying Copilot?
- Copilot is a tool. We teach your people to delegate to it, one task at a time, with the brief, the check and the sign-off that make it safe to use.
Begin
Start with one task.
Thirty minutes about your week. We name the first task worth handing to an agent.
30 minutes · no deck · we say plainly if we are not the right fit