Two years ago, the work sitting on my desk right now would’ve required six or seven people. Not hypothetically. Not in some LinkedIn thought-experiment sense. Literally.
Managing system integrator partnerships across multiple countries for an open-source civic registration platform. Running B2B corporate training expansion across the GCC. Delivering AI strategy work for enterprise clients. Building out partnership frameworks for government programmes. Writing proposals, competitor analyses, briefing packs, market scoping documents. The full works.
That used to be a team. Business analysts. Project managers. A researcher or two. Someone on decks. Someone on comms. Maybe a junior doing the financial modelling while a senior reviewed it and a partner took the credit. A proper squad, billing collectively at a rate that would make your eyes water.
Now it’s me. One person. A flat in Dubai. A dog who couldn’t care less about consulting margins. And an AI stack. A few well-configured agents, intelligent model routing, and twenty-odd years of knowing which questions actually matter and which ones are just noise dressed up in a slide deck.
I didn’t plan this. I wasn’t sitting in some co-working space with a whiteboard full of sticky notes thinking “right, I’m going to disrupt the consulting industry.” I was just trying to survive running multiple roles simultaneously without my brain catching fire.
But somewhere in the last 24 months, something shifted. The AI agents stopped being fancy autocomplete and started being the team. Research, synthesis, first drafts, competitive analysis, document generation, daily intelligence briefings. All handled. Not perfectly. But at 90% of what a decent junior consultant would produce, delivered in minutes instead of days.
One person. A few AI agents. And the output of a small consultancy. That’s not a pitch. That’s my Tuesday.
Two years. That’s all it took for the maths to completely break.
The Body Count Model (Or: Why Consulting Was Always a Staffing Business in a Nice Suit)
Here’s something nobody in consulting wants to admit out loud: the traditional model never really sold expertise. It sold bodies. The partner won the deal. Brilliant at dinners, golf, and the subtle art of making a £250K proposal sound like a bargain. The senior manager scoped the work. And then a squad of junior analysts, fresh out of business school, still excited about Excel shortcuts, billing at £800 a day, did the actual graft.
Market research. Competitor analysis. Slide decks with so many charts they could double as wallpaper. Financial models that took three weeks to build and forty-five minutes to present. The client paid for the firm’s logo, the partner‘s handshake, and the comforting heft of a 120-page PDF that said “we hired serious people to think about this.”
The uncomfortable maths? On a Big 4 engagement for something like “AI strategy” (and I’ve sat on both sides of these, so this isn’t guesswork) you might be looking at £150K to £300K. Of that, maybe 15% is genuine senior thinking. The kind of insight that actually changes a decision. The rest is research, formatting, project management overhead, and enough junior billing hours to keep the leverage ratio healthy.
Right. Before this turns into one of those “I cracked the code and now I’ll sell you a course for £2,997” posts. Let me be clear. I didn’t crack anything. The tools did. I just happened to be paying attention while most people were still arguing about whether ChatGPT could write a decent email.
And the data backs this up. Virtasant surveyed 384 consultants and found that AI is saving them roughly 13 hours per week while slashing project costs by up to 90%. Not 10%. Not “a bit cheaper.” Ninety percent. Their research suggests that a two-person AI-augmented team can now produce what used to require twenty people. Let that sink in for a second.
The Moment the Maths Stopped Making Sense (For Them)
The moat for big consulting firms used to be built on three things: access to data, proprietary frameworks, and a deep bench of smart people willing to work stupid hours. AI has blown through all three.
Data? Every decent model on the market can now synthesise research from thousands of sources in minutes. The competitor analysis that took a junior analyst two weeks and seven Red Bulls? An agent does it before my morning coffee gets cold.
Frameworks? Porter’s Five Forces doesn’t require a McKinsey badge to apply. It never did. The frameworks were always public. What the firms sold was the confidence to apply them. And a logo that told the board “we’ve been rigorous about this.”
Talent? The junior analyst doing the grunt work. The research, the data cleaning, the first-draft everything. Automated out of the equation. Not partially. Not “augmented.” The actual tasks those people used to do are now performed by software that costs less per month than their daily sandwich budget.
So here’s the question nobody in a Big 4 boardroom wants to answer: if the deliverable is the same quality, why is the client paying for fifteen people when one person with a properly configured AI stack can do it? They’re not. Not for much longer, anyway.
What Actually Changed (And It Wasn’t ChatGPT)
Look, the shift here isn’t about having access to a large language model. Everyone has that now. Your nan has that. She’s probably using it to write passive-aggressive messages to the parish council.
The actual shift, the one that matters, is the cost of intelligence labour dropping to near-zero. The stuff that used to require warm bodies and billable hours: research, synthesis, first drafts, data analysis, competitive scans, market sizing, stakeholder mapping. All of it effectively free now.
I run Jarvis, a memory-persistent AI agent I’ve hashed together from open-source components that anyone could find with a good old ChatGPT search. Nothing proprietary. Nothing fancy. Just well-configured pieces bolted together with a clear routing logic I call Protocol B.
Opus 4.6 handles the complex reasoning and heavyweight strategy work. MiniMax 2.5 does the bulk of everything else. And Kimi 2.5 picks up the grunt work: research, summarisation, the tasks that need volume more than brilliance.
Brought my monthly API costs down from $870 to under $200. Under $200 a month. That’s the infrastructure cost of running what is, functionally, a one-person consulting operation producing work that used to require a team and a half. That’s cheaper than a WeWork hot desk. It’s cheaper than the coffee budget at most agencies. And it runs 24/7 without needing a pep talk or a performance review.
The Moat That Actually Matters (Spoiler: It’s Not the Tech)
So if everyone has the same AI tools (and they do, or they will within six months) what’s the actual differentiator? It’s not the tech. It’s the twenty years that came before the tech. The moat is knowing which questions to ask in the first place. It’s having sat in enough boardrooms to know when a CEO is telling you what they think you want to hear instead of what’s actually happening. It’s the pattern recognition that comes from having delivered fifty projects across healthcare, government, financial services, and emerging markets. And knowing that the thing that killed the last three isn’t in the data. It’s in the politics.
AI handles the “what.” It can pull the research, build the model, draft the analysis, and format the deck faster than any human team. But it can’t do the “so what” and the “now what.” A junior analyst with Claude can produce a brilliant competitor analysis. Genuinely impressive output. But they can’t sit across from a CEO and say: “This section is noise. Here’s the one thing that matters. And here’s what you should do about it by Friday.” That takes judgement. Judgement takes years. There’s no shortcut, no prompt template, no bloody “AI mastery course” that replaces it.
That’s it. That’s the whole thing.
The Big 4 Know This Is Happening (And They’re Bricking It)
Bain was the first major consulting firm to partner with OpenAI. McKinsey built QuantumBlack, their dedicated AI division. Deloitte and KPMG are pouring money into AI practices faster than they can hire for them. Sia, a mid-tier consultancy, just launched an “Agent Store” with over 400 AI agents that act as analysts, strategists, and auditors across every corporate function.
Now, you might look at all that and think: “See? The big firms are adapting. They’ll be fine.” Bollocks. They’re not building AI practices to sell AI to clients. Not primarily. They’re building them to protect their own margins before the economics collapse underneath them.
The consulting model runs on leverage ratios. One partner. Eight juniors. The partner wins the work, the juniors deliver it, and the difference between what the juniors cost and what the client pays is where the profit lives. When AI replaces six of those eight juniors, and it’s getting there fast, the entire financial model falls apart.
Gartner reckons 40% of enterprise applications will have AI agents embedded by the end of 2026. Up from less than 5% in 2025. That’s not a slow transition. That’s a sprint. And the infrastructure clients need to do this stuff themselves, or to hire one experienced person with a good stack instead of a team of twenty, is maturing at a pace that should terrify every mid-tier consultancy without a genuine specialism.
The 90% Cost Drop Nobody’s Talking About
Right. Let’s talk money, because that’s where this gets properly uncomfortable for the old guard. Virtasant’s data shows project costs dropping by up to 90% when AI is properly integrated into consulting delivery. Not “efficiency gains.” Not “modest savings.” A 90% reduction.
In practice, that means a strategy engagement that used to cost £200K can now be delivered for £20K to £30K by a solo operator with deep domain expertise and a well-built AI stack. The client gets faster turnaround, direct access to the person actually doing the thinking (not a junior analyst relaying messages from a senior who’s too busy to be on the call), and a deliverable that’s 95% as polished as the Big 4 version.
The 5% they lose? The logo on the cover page and the ability to tell the board “we hired McKinsey.” That 5% is expensive insurance. But it’s getting harder to justify when the alternative costs a tenth of the price and the person on the call actually knows your industry. Because they’ve spent two decades in it, not two months Googling it.
Who Wins in This New World (And Who’s Properly Screwed)
The winners are domain experts. People with ten, fifteen, twenty-plus years of real experience, in real industries, with real clients, having made real mistakes, who learn to wield AI as infrastructure rather than a novelty. People who’ve been in the rooms, done the deals, survived the politics.
Also winning: small, specialist firms that move fast, price on value instead of hours, and don’t carry the overhead of a hundred juniors who now have nothing to do. Here’s an interesting one from the Virtasant data: female consultants are 30% more likely to be ahead of the curve on creative AI strategy problem-solving. The people winning this transition aren’t necessarily the ones you’d expect. It’s not the loudest voices on LinkedIn. It’s the people quietly getting shit done with better tools.
And the losers? Mid-tier consultancies without strong brands or deep specialisms. The “body shop” model, “we’ll send you eight warm bodies at £600 a day each,” is toast. If your value proposition was “we have lots of people,” you’re about to discover that “lots of people” is no longer a selling point when one person and a few agents can match the output.
My Stack (Because Someone’s Going to Ask)
No guru energy here. No “buy my framework.” Just what I actually use and what it costs.
Jarvis is a memory-persistent AI agent. Not a product. Not a platform. A thing I’ve stitched together from open-source components, stuff anyone could find with a bit of searching and some patience. It lives on a VPS, talks to me via Telegram, and runs Protocol B for model routing.
Opus 4.6 handles the complex reasoning and strategy work. MiniMax 2.5 does the bulk of everything else. And Kimi 2.5 picks up the grunt work: research sweeps, summarisation, the high-volume tasks that need speed more than sophistication.
Daily intelligence briefings are automated. Document generation, competitive analysis, research synthesis, proposal drafting. All agent-handled with my review and direction. Monthly cost: under $200 in API spend. VPS hosting is negligible.
The previous equivalent in human hours, if I’d hired contractors or a small team to do what Jarvis now handles, would’ve easily run £5K to £8K a month. And it wouldn’t have been available at 2am when I’m working across time zones.
But here’s the thing, and I’ll keep saying this until I’m blue in the face: the point isn’t the tools. The tools change every six months. What I’m using today will probably look quaint by Christmas. The point is that the barrier to running a high-output consulting practice has dropped from “you need a firm” to “you need a laptop and a decade of knowing what you’re doing.” The tools are the easy part. The decade is the hard part.
The Bit Nobody Wants to Hear
Right. Before anyone mistakes this for one of those “quit your job and buy my $97 ebook” stories that LinkedIn is drowning in — let me be straight about two things.
First: I didn’t plan any of this. There was no strategy. No vision board. No “five-year plan to disrupt consulting.” I was juggling three roles, trying to keep my head above water, and the AI stack wasn’t a grand design. It was survival. Sod it, let’s automate this before I lose my mind. But somewhere along the way, the output started looking like what a mid-tier consultancy would charge six figures to produce. And I thought: hang on.
Second, and this is the bit that’ll annoy the “anyone can do this” crowd: this doesn’t work for everyone. If you’re 22, your only experience is a summer internship and a Udemy course, and your plan is to call yourself an “AI strategy consultant” — AI doesn’t magically give you twenty years of judgement. The tools amplify what you already know. If you don’t know much yet, they amplify that too. Quite faithfully, in fact.
But. If you’ve got the domain expertise. If you’ve done the years. If you’ve sat in the rooms and run the programmes and cleaned up the messes. And you’re still billing by the hour, or working inside someone else’s leverage model, watching partners take credit for your thinking while you get a pat on the back and a 3% raise. The question isn’t whether to make the jump. It’s why you haven’t already.
Your move.
If this landed, the Sunday Blueprint is where I write about this stuff every week. Real insights, zero hype, none of that inspirational LinkedIn bollocks. Just a bloke in Dubai trying to figure this out in real time and sharing what’s actually working.
Next week: the actual nuts and bolts — how a solo AI-augmented consultant delivers a strategy engagement from pitch to final deliverable. The workflows, the prompts, the bits that go wrong.
P.S. If you’re a Big 4 partner reading this and thinking “this bloke’s delusional” — fair enough. But check your utilisation rates for Q4 and then come back to me.
