What an AI consultant actually does, beyond the prompts

Half the work is unprompting.

When clients hire an AI consultant they have a list. Automate the inbox. Generate the proposals. Summarise the calls. The list is usually wrong. Not because the items are wrong, but because the items in front of them are the visible ones. The work that's worth automating, or worth rebuilding entirely, is usually one layer underneath.

What an AI consultant actually does

Three things, mostly.

One, they tell clients what not to automate. Plenty of AI consulting is the negative space: pointing at a workflow and saying it's already good enough, or saying the cost of automating it exceeds the benefit because the workflow only runs four times a month. A consultant who can't tell you not to automate something is selling you software, not advice.

Two, they find the rebuild. The real opportunity is rarely "make this faster". It's "this whole pipeline exists because we didn't have AI when we built it". The proposal generation that takes a sales rep two hours per deal exists because nobody could read a brief, find the right case studies, and assemble a draft, all at once. Now they can. The fix isn't faster proposal generation. The fix is: proposals are now a 60-second job, so what work upstream of them can change because of that?

Three, they ship the thing. Strategy without shipping is theatre. If the deliverable is a deck, you've hired the wrong person. The deliverable is the running system, with the prompts version-controlled, the failure modes documented, and the on-call rota agreed.

What AI consulting is not

It's not "I know how to prompt Claude". Anyone with a free Claude account knows that.

It's not "here are the best AI tools right now". That's a newsletter, not a consultancy.

It's not even "I'll build you a chatbot". A chatbot is an output. The consulting is the bit that decides whether you should have one.

Why Leeds, why now

I'm an AI consultant working from Leeds. The location matters because most of the AI consulting market is concentrated in London, San Francisco, and a handful of US hubs. UK businesses outside that orbit have to either pay London rates with London travel, fly someone in from California, or hire a generalist consultant who'll learn AI on their dime.

There's a third option, which is to work with someone who's already AI-native, ships products on Claude every day, and lives 200 miles up the M1. That's the gap I sit in.

The work I'd do for you

A typical engagement looks like this.

1. A two-day audit. I sit with your team, look at the pipelines, watch the work happen. No slides. Just observation and a notebook.
2. A written brief. What to automate. What to rebuild. What to leave alone. Honest about the parts where I'd want to pilot before committing.
3. A delivery phase. Usually four to twelve weeks, depending on scope. Working software at the end, not a strategy document.
4. A handover. Prompts in a repo, agents in .claude/, runbooks for the on-call team, the patterns documented so your team can extend it.

What I won't do

I won't ghostwrite your AI strategy doc so you can hand it to procurement. I won't run prompt engineering training sessions for fifty people. I won't sell you a tool stack I haven't shipped on myself.

If you want a Claude consultant in the UK who's actually building, drop me a message. I'm working out of a one-bedroom office in Leeds and the only diary I keep is the one for shipping.

Source: /content/articles/what-an-ai-consultant-actually-does.mdx

More writing → /writing. Got a brief? Get in touch.