AI consultant vs AI agency: how UK mid-market operators should choose

The two models, plainly

An AI consultant is typically a senior engineer running their own practice (often founder-led, often single-person or a tight pair). They run discovery, write the scope, design the system, write the code, deploy it, and own the result end-to-end. You talk to them directly throughout. No account manager, no project manager layer, no internal handoffs between team members.

An AI agency is a multi-disciplinary firm — usually 10 to 100+ people — with account managers, project managers, strategists, designers, and engineers across separate roles. The work moves through a handoff chain: account manager scopes it, PM runs it, strategist plans it, engineers build it, you receive progress reports. Decisions are made in internal meetings before they reach you.

Both models are legitimate. The right choice for UK mid-market AI builds (typically £25k–£200k) is almost always the consultant. Above £500k of scope, the picture shifts. The five criteria below explain why.

The 5-criterion comparison

Decision flowchart

For your build, answer "consultant" or "agency" for each criterion:

1. Scope size — is the build under £200k? (Almost all UK mid-market AI builds are.)
2. Speed and decision cadence — do you need to ship in weeks, not quarters?
3. Code ownership and risk profile — do you want to own the code with no vendor lock-in?
4. Technical depth needed — does the build need deep engineering judgement, or broad parallel execution?
5. Discovery quality — does the spec need to emerge from a conversation, or is it already written?

Three or more "consultant" answers and you should be hiring a consultant. Four or five and the agency path is almost certainly the wrong choice. Three or more "agency" answers — typically only for enterprise-scale rollouts — and an agency is the right fit.

Cost framing — what the day-rate difference actually means

Honest 2026 UK day-rates:

• Senior AI engineer at a consultancy (one-person or tight team): £800–£1,400/day. You\'re paying for senior engineering time directly.
• Senior AI engineer at a UK agency: £1,200–£2,000/day. Same engineer, plus a billable layer of account management, project management, and overhead.
• Junior engineer at a UK agency (the person who often actually does the work): £600–£1,000/day billed; £200–£350/day true salary cost. The agency margin sits between those numbers.
• Offshore agency engineer: £200–£500/day billed. Lower technical depth in most cases; coordination overhead eats much of the saving on UK mid-market discovery-led work.

For a typical £80k UK mid-market AI build (8–12 weeks of senior engineering plus integration work), the consultant model usually delivers it at the bottom of the range; the agency model usually delivers similar scope at 1.5–2x the cost because of the overhead layered around the same engineering effort.

When a consultant is the right answer

• Build size is under £200k.
• You want to talk directly to the engineer doing the work.
• You need to ship in weeks, not quarters.
• You want code in your repository, not on an agency server.
• You value discovery quality — the spec emerging from a technical conversation.
• Your operation is a UK mid-market business — founder-led or with a tight ops team.

When an agency is the right answer

• Build size is £500k+ with genuinely parallel workstreams.
• You\'re in a regulated enterprise where formal process is part of the deliverable.
• The spec is already written and the work is well-defined and bid-ready.
• Your procurement team can\'t contract with single-engineer practices for legal or insurance reasons.
• You explicitly want the risk-diversification of multiple engineers on the project (i.e. truck-factor protection).

The five tests that filter genuine consultants from prompt-slingers

The biggest risk when hiring an AI consultant isn\'t the consultant model — it\'s hiring the wrong consultant. Five concrete tests, covered in detail in the how to evaluate an AI consultant guide:

1. Ask them to show production code from a live build (not a demo, not a screenshot).
2. Ask them to walk you through how they\'d handle data sensitivity for your specific data.
3. Ask them what they\'d tell you NOT to build.
4. Ask for the integration list on their last build (and what was hard about it).
5. Ask whether you\'ll own the code at the end.

A prompt-slinger fails one or more of these in the first ten minutes of conversation. A real engineer-led consultant answers all five clearly.

What I would tell you on a discovery call

On a discovery call, you\'re talking directly to me — Dean — the engineer who would do the work. No account manager. No PM. No internal meeting between us and the actual build. That\'s the whole consultant model expressed in one mechanism. If your build looks like the £500k+ enterprise rollout the framework above describes, I\'ll tell you on the call and point you at agencies who fit better. If it looks like a UK mid-market build, you\'ll leave with a costed bottleneck map and a clear next step.