Book a fit call
Book the free 30-minute fit call

Or call 0161 543 7189

Co-founder, one of two

Sam Wall

Architecture, AI behaviour and what is safe to build

Sam Wall is co-founder of AGMM, a UK AI consultancy. While 95% of the work runs on the systems the business has built, Sam spends his time on the 5% that is judgement: architecture, what is safe to automate, how AI behaves, and whether a build will hold up once people depend on it.

By Sam Wall and Adam Gillett · Checked on

Sam Wall, co-founder of AGMM, an AI consultancy
Sam Wall, co-founder, AGMM.Decides what is built, bought, connected or left alone.

Where his time goes

95% of AGMM runs without Sam. He keeps the 5% that decides.

That is the design, and it is why the business scales. The content line, the outreach engines, the model fleet and the operating brain carry the volume, department by department. What is left for Sam is the judgement with the most leverage: the calls where a wrong answer is expensive and a right one pays for everything around it.

95%

of AGMM runs without Sam. The systems carry it, every day, at a volume a large team would struggle with.

5%

is Sam’s: the calls that decide whether a build pays.

  1. Client builds
  2. Proposals and pricing
  3. Running client systems
  4. Outreach engines
  5. Content line
  6. Autoposter
  7. Model fleet
  8. Quality gates and review
  9. The operating brain
  10. Research
  11. Finance and admin
  12. Library and learning

Sam’s 5%

Build, buy or connect

Which of a client’s tools are worth keeping, and what needs building.

Where a model may decide

Where AI earns its place, what it is allowed to decide, and what a person keeps.

When it is ready to ship

Tested against the client’s real data and exceptions, then released without stopping the business.

Pricing the big build

Scope, price and phasing on the proposals that matter most.

A week at AGMM, illustrated. Each square stands for five pieces of work done by our systems. Gold squares are the calls Sam keeps. Volumes illustrative.

The 5%, in a calendar

What that slice looks like across a week.

Client sessions, build reviews and the decisions that settle what gets built first. Every meeting has a small agent sitting in, taking notes and filing what we learned, so the next build starts further ahead.

  1. Monday, 08:30Founders: the week aheadDecided: What gets built first
  2. Monday, 15:30Architecture decisionDecided: Connect, do not rebuild
  3. Tuesday, 13:00Build review: schedulingDecided: Exceptions go to a person
  4. Wednesday, 08:30Proposal walk-throughDecided: Phase one scope agreed
  5. Wednesday, 15:00Founders: pricingDecided: The Audit stays £2,000
  6. Thursday, 11:00Model choice reviewDecided: Long files, bigger model
  7. Thursday, 13:00Audit read-outDecided: Ranked plan signed off
  8. Thursday, 16:45Release: go or no-goDecided: Ship on Friday
  9. Friday, 08:30Founders: the week behindDecided: Retire two skills, add one
  10. Friday, 15:30Week close: decisions logDecided: Next week's priorities
MondayTuesdayWednesdayThursdayFriday
  1. Founders: the week aheadDecided: What gets built first
  2. Fit call
  3. Audit session 1 of 3
  4. Build review
  5. Fit call
  6. Architecture decisionDecided: Connect, do not rebuild
  1. Audit session 2 of 3
  2. Fit call
  3. Data walk-through
  4. Build review: schedulingDecided: Exceptions go to a person
  5. Handover rehearsal
  1. Proposal walk-throughDecided: Phase one scope agreed
  2. Fit call
  3. Audit session 3 of 3
  4. Integration call
  5. Founders: pricingDecided: The Audit stays £2,000
  6. Build review
  1. Fit call
  2. Discovery session
  3. Model choice reviewDecided: Long files, bigger model
  4. Audit read-outDecided: Ranked plan signed off
  5. Fit call
  6. Release: go or no-goDecided: Ship on Friday
  1. Founders: the week behindDecided: Retire two skills, add one
  2. Fit call
  3. Deployment check-in
  4. Build review
  5. Week close: decisions logDecided: Next week's priorities

Behind every week, the systems keep running:

  • Content line, up to 15 short-form videos and one YouTube film a day.
  • Autoposter, up to 10 posts a day, five platforms, receipts kept.
  • Outreach engines, Instagram and LinkedIn, ramped caps.
  • Model fleet, every job routed, every output reviewed.
  • The brain, what each meeting taught us, filed.

The small mark on each meeting: an agent sits in, takes notes, and files what we learned.

A typical week in Sam's calendar. The meetings are representative; the systems running behind them are real.Demonstration of our real architecture. Timings compressed.

What he decides

What does Sam decide?

For you, it means the person deciding your architecture is the person in the room, not a name on a slide. These are the calls Sam keeps.

  1. Build, buy or connect

    Which of your existing tools are worth keeping, and what needs building.

    For exampleKeep the accounts package, connect the bank feed and build only the matching rules.

  2. Where a model may decide

    Where AI earns its place, what it is allowed to decide, and what a person keeps.

    For exampleA model may draft the supplier chaser. A person approves any credit decision.

  3. When it is ready to ship

    Tested against your real data and exceptions, then released without stopping the business.

    For exampleThe new month-end runs beside the old one for a full close before it replaces it.

  4. What the system learns

    What each week taught us, filed so the next build starts further ahead.

    For exampleAn exception that had to reach a person twice becomes a rule the system applies itself.

What he builds

What does Sam build?

The system underneath the promise. When AGMM agrees to remove an operating constraint, somebody has to decide how the data will be shaped, which of the existing tools are worth keeping, where a model may be used and where it must not be, what happens on the day a supplier changes a file format, and how the thing gets into production without stopping the business for a week. That is Sam's half of the work.

In practice that means:

  • The architecture: what is built, what is bought, what is connected and what is left alone.
  • The build decisions, including how the system connects to software a business already pays for.
  • AI behaviour: where a model earns its place, what it is allowed to decide and what a person keeps.
  • Testing against real data, real rules and the exceptions your team already works around.
  • Deployment, and the handover that leaves your team able to run the system without AGMM in the room.

Adam Gillett owns the other half: the process, the people and the adoption. Neither half works on its own. That is why AGMM sells AI consultancy that does not stop at a document: the founders who diagnose a problem stay to see the system running in ordinary use.

How he runs the audit

How does he run the £2,000 audit?

The AI Constraint Audit is fourteen days and three sixty-minute working sessions. Sam is in all three, and the sessions are the smaller part of it. His job between them is to test whether each candidate opportunity can be built against the systems and the data the business really has, rather than the systems and the data it believes it has.

That means asking for the awkward things early: the export nobody can produce, the field three departments fill in differently, the rule that lives in one person's head. An opportunity that cannot survive those questions does not reach the economic cases, because a business case built on a dependency that does not exist is a way of losing money slowly.

The audit fee is not deducted from a build, so there is no financial reason for Sam to arrive at a recommendation to build. Where the answer is a process change, an integration or nothing at all, that is the answer that gets written down, and you keep the four documents either way. A fuller audit always follows inside any build proposal.

What it runs on

Inside the systems he built for AGMM

We built these systems 6 to 12 months before off-the-shelf tools existed, and they run AGMM every day. One of our knowledge bases holds 4,535 notes, 5.9 million words and 1,023 skills. Around it sit the model fleet, the outreach engines, a content line that is one small part of marketing, and an autoposter that keeps a platform receipt for every post.

Sam owns how they are put together and where they stop: which model does which job, what a reviewer must sign off, and what is never automated. Under the hood it is folders, files and links, organised properly. Every department of the business is drawn on Inside AGMM, and the brain itself is explained in this guide.

Verification

How to check any of this

The company
AGMM LTD, company number 16384156, incorporated 13 April 2025. Read the Companies House record.
The systems
What runs inside AGMM while the founders are in their 5% is drawn, department by department, on Inside AGMM.
The price
£2,000 for the fourteen-day audit, published on what it costs and compared with other UK firms on the cost comparison.
Your data
What AGMM does with client information during an audit and a build is written down on security and data protection.
Sam directly
LinkedIn, or 0161 543 7189.

Thirty minutes with the person who would build it.

The fit call is free, nothing is bought on it, and it is Sam and Adam on the call rather than a salesperson. If the answer is that AGMM is the wrong firm for the problem, you get that answer there.

Book the free fit callCall 0161 543 7189

The Audit: £2,000, 14 days, four outputs you own.

  1. MessyPacks built by hand the night before
  2. OrganisedOne set of figures every Monday
  3. RunningPacks assemble themselves; you decide
Lit: Leadership and decisions, where the call starts. The audit then looks across all twelve departments for constraints, weak points and money to save or earn.