In December 2025, McKinsey announced plans to deploy AI agents at a scale matching its human workforce — 40,000 agents by end of 2026.[^1] Accenture committed $3 billion to AI investment and began embedding generative AI across every client engagement.[^2] Deloitte, PwC, EY — all followed with their own AI-first operating model announcements within weeks of each other.
The market response was overwhelmingly positive. Analysts praised the efficiency gains. Clients anticipated lower fees and faster delivery. The consensus narrative was simple: AI makes consultants better, which makes consulting better, which makes your transformation outcomes better.
That narrative is incomplete. And the part it leaves out is the part that matters most.
The Productivity Asymmetry
The consulting firms are not exaggerating their AI gains. The numbers are real.
BCG published internal research in late 2025 showing that consultants using AI tools completed strategic analysis 40% faster with measurably higher quality.[^3] McKinsey reported similar productivity improvements across its knowledge work — research, synthesis, document production, code generation. Deloitte's own data showed AI-augmented teams delivering client work in roughly half the time of their non-augmented equivalents.
These are genuine operational improvements. We are not disputing them.
But look at the shape of the improvement. The consultant gets faster. The consultant's firm captures margin. The client receives the deliverable and then — manually reviews it, manually integrates it, manually operationalises it — at exactly the same speed as before.
This is the productivity asymmetry. One side of the engagement is accelerating. The other is standing still.
A CIO we spoke with in January put it plainly: "They delivered the architecture review in three weeks instead of eight. Brilliant. Then it sat on my team's desk for four months because we didn't have the internal capability to evaluate it, let alone implement it." The bottleneck was never on the consulting side. It was on the client side. And the consulting firm's AI made the gap wider, not narrower.
Speed Without Absorption Is Just Faster Dependency
There is a compounding effect here that deserves attention.
When a consulting engagement delivers faster, the natural response is to commission more work. The logic is seductive — if they can do an eight-week sprint in three, we can run three sprints instead of one this quarter. More velocity. More output. More progress.
Except that "progress" is measured in deliverables produced, not capability transferred. Each sprint generates more intellectual property that lives in the consulting firm's systems, more proprietary tooling trained on your data, more institutional knowledge held by people who do not work for you.
Gartner's 2026 projections are worth noting here. They predict 40% of agentic AI projects will be cancelled or scaled back by 2027 — not because the AI failed, but because organisations lacked the internal architecture to operate it.[^4] The technology worked. The absorption didn't.
Every AI-augmented engagement that succeeds on its own terms reinforces a structural pattern: the consulting firm builds capability. You consume output. The gap between what they can do and what you can do — independently — grows with each successful project.
This is not malice. It is mechanics.
40,000 Agents, Zero of Them Yours
McKinsey's 40,000-agent target is worth examining closely, because it reveals the strategic direction of the entire consulting industry.
Those agents are not general-purpose productivity tools. They are trained on McKinsey's proprietary methodologies. They encode decades of engagement patterns, industry benchmarks, and delivery playbooks. They represent the firm's accumulated intellectual capital — automated, scalable, and deployable across thousands of client engagements simultaneously.
This is a genuine competitive advantage for the firm. It is also, by design, a competitive advantage that cannot transfer to the client.
When the engagement ends, the agents leave. The models leave. The tooling leaves. What remains is a set of deliverables — PDFs, slide decks, architecture diagrams, roadmaps — that represent the output of capability, not the capability itself.
Compare this to what happens when an organisation builds its own AI capability. The models stay. The training data stays. The institutional knowledge encoded in the system stays. Every engagement with your own tooling makes the tooling better — and that improvement compounds internally rather than externally.
The distinction is not theoretical. It is the difference between renting compute and owning infrastructure. Between licensing software and building product. Between hiring a driver and learning to drive.
The Strategic Choice Nobody Is Framing Correctly
The conversation in most boardrooms has been reduced to a binary: adopt AI or fall behind. That framing is already obsolete. Adoption is table stakes. Deloitte's 2026 survey found that two-thirds of business leaders say AI is advancing faster than their organisation can absorb it.[^5] The adoption question is settled. The absorption question is not.
The real strategic choice is between two models of AI capability:
Rented capability. You engage firms whose AI tools augment their delivery. You receive better outputs, faster. You pay for access to their capability on a per-engagement basis. When the engagement ends, the capability leaves.
Owned capability. You invest in building AI systems that serve your organisation directly. The models are trained on your processes. The agents understand your operations. When a vendor relationship changes, your capability persists.
Most enterprises are defaulting to the first model without recognising it as a choice. The consulting firms' AI announcements made rented capability the path of least resistance — faster procurement, lower upfront investment, immediate productivity gains. The second model requires deliberate investment in internal architecture, talent, and governance that takes longer to build and is harder to justify in a quarterly review.
But the cost curves tell a different story over time.
Token pricing for large language models has dropped by a factor of 1,200 — from roughly $600 per million tokens to $0.50.[^6] The raw technology has never been cheaper. What remains expensive is the organisational capability to use it — and that expense is either an investment (if you're building internally) or a recurring fee (if you're renting externally).
The Consultant Productivity Paradox
Here is the counterintuitive finding buried in all of this.
The better consulting firms get at using AI, the more critical it becomes for clients to build their own AI transformation capability. Not despite the productivity gains. Because of them.
When a consultant could produce one strategy document per week, the client had time to absorb it. Review it. Pressure-test it against internal knowledge. Push back on assumptions. Build understanding.
When that same consultant produces three documents per week — each more comprehensive than before — the client's absorption bandwidth hasn't changed. The review bottleneck tightens. The gap between what is delivered and what is understood widens. Decisions get made on the basis of outputs that nobody inside the organisation fully evaluated, because there wasn't time, and because the consulting firm's AI moved faster than the client's human process.
PwC's 2026 Global CEO Survey found that 56% of CEOs report zero financial payoff from AI investments.[^7] Not "disappointing returns." Zero. That number makes more sense when you understand the absorption problem. The AI worked. The organisation couldn't metabolise what it produced.
This is not an argument against using consultants. Consultants bring expertise, external perspective, and surge capacity that most organisations need at various points. The argument is against using consultants as a substitute for building the capability to evaluate, absorb, and extend what they deliver. Those are different functions. Conflating them is what creates the dependency spiral.
What This Means for the CIO
If you are a CIO reviewing your consulting portfolio right now, three things are probably true.
First, your consulting partners are delivering faster than they were twelve months ago. Their AI-augmented teams are genuinely more productive. The outputs are better. The timelines are shorter. This is real.
Second, your internal teams' capacity to absorb those outputs has not increased at the same rate. You have more deliverables landing on fewer desks, reviewed by people who are already running at capacity on operational work. The absorption bottleneck is widening.
Third, nobody in the organisation is explicitly tracking the gap between what is being delivered and what is being absorbed. There is no metric for it. No dashboard. No steering committee agenda item. The gap is growing in silence.
The AI shift that December 2025 announced is real. Consulting firms will be faster, more capable, more scalable. That is not the threat. The threat is that the asymmetry between their capability and yours becomes structural — not because anyone designed it that way, but because nobody designed against it.
Your Waypoint
Pull your current roster of consulting engagements — every active statement of work. For each one, answer this question: what AI capability is being used by the consulting team that your internal team cannot replicate independently?
Make two columns. Left column: AI capabilities the consultants are using in this engagement. Right column: AI capabilities your organisation currently owns and operates without external support.
The gap between those columns is your dependency exposure. Every item on the left that has no equivalent on the right represents capability you are renting, not building. Some of that rental is the right strategic choice — you may not need in-house expertise in every domain. But if you cannot point to the decision where that was deliberated and documented, the dependency is accidental, not strategic.
For your next engagement renewal or new consulting procurement, add one clause: a capability transfer plan. What specific AI capability will your team be able to operate independently after the engagement ends? If the answer is "none" — and it will be, more often than you expect — you have identified the structural problem.
The TCM Crisis Report examines the full scope of this capability transfer gap, including the evidence base behind why 88% of enterprises have adopted AI while only 6% are capturing value from it.
[^1]: McKinsey Global Institute, internal communications and press briefing, December 2025. [^2]: Accenture, "AI Investment and Transformation Strategy," investor presentation, Q4 2025. [^3]: Boston Consulting Group, "The Impact of AI on Knowledge Work," research publication, 2025. [^4]: Gartner, "Predicts 2026: Agentic AI," analyst report, Q1 2026. [^5]: Deloitte, "State of Generative AI in the Enterprise," Q1 2026, survey of 3,235 business leaders. [^6]: Industry composite pricing data. GPT-3 launch pricing (2020) vs. current frontier model API pricing (early 2026). [^7]: PwC, "27th Annual Global CEO Survey," 2026, survey of 4,454 CEOs.