Watch the full episode on YouTube: https://youtu.be/bT39EqfzIYw
The intelligence core is Usman Sheikh’s answer to the hardest question in professional services: what happens to consulting firms when AI dissolves the pyramid that built them? In this episode of The Innovation Show, the founder of High Output Ventures and author of the Framebreak newsletter joins Aidan McCullen to map the shift from pyramids to rings — and what it means for talent, margins and the future of knowledge work.
Why firms became pyramids
Professional service firms solved a scarcity problem. A firm’s most valuable asset was its senior leaders’ judgment, but there were only so many hours in a partner’s day. So firms built layers of managers and juniors beneath the partners to scale that judgment across far more clients than one person could serve.
The pyramid did three jobs at once. First, it leveraged senior judgment. Second, it generated margin — the delta between what the firm charged for junior work and what it paid for it. Third, and least discussed, it was a training ground: through years of supervised error correction, juniors became the next generation of “signers” who could put their name on a document. Best of all, the client paid for that training through the billable hour. AI, Usman argues, pulls on all three threads at once.
Judgment-shaped artifacts: why AI is not offshoring 2.0
Past shifts left the pyramid intact. Offshoring changed where the work got done; workflow software and RPA changed how predictable processes ran. AI, by contrast, changes who does the work — because it produces what Usman calls judgment-shaped artifacts. The output looks like judgment, but it contains none. Someone who has never written a brief will think the machine wrote a great one; someone who writes briefs for a living will spot the four places it went wrong.
That gap is dangerous for juniors. In his essay “Fluent and Substitutable,” Usman describes a junior who kept pressing the button hoping the LLM would land on the right answer — turning the model into a slot machine. His advice: do the work by hand until you can grade the machine’s output. Because if you and the machine are both guessing, the machine gets more reps, and the machine wins.
What is the intelligence core?
The intelligence core is a feedback loop. The machine produces an output; someone with genuine expertise error-corrects it; the work goes into the world; the results come back; and every correction feeds the core, making the next output better. Instead of logging the winning ad, contract or model in a static template library, the firm compounds its learning inside the system itself.
Once the intelligence core works, leverage no longer comes from adding people. As a result, the firm can morph from a pyramid into what Usman calls a ring — expertise arranged around a learning centre. Crucially, this is not a digital twin. A digital twin is a snapshot of one leader’s mind that decays quickly; the core is a living flow of the firm’s ongoing work.
The AI discount and where the surplus drains
If AI creates a surplus, who captures it? Clients are already asking — KPMG reportedly requested an AI discount from its own auditor. Usman sees the surplus draining in two directions. It drains to clients wherever work is low-stakes and easily checkable. And it drains to platforms, which extract rent on every token or loop — which is why Palantir’s and Microsoft’s leaders are suddenly preaching sovereignty over data and models.
Where results are slow to verify and the cost of being wrong is high — market-entry strategy, Fortune 500 audits — the drain is slower. Your position on that spectrum decides how much surplus you keep, and whether you can build the “locks” (Usman’s smaller, sharper successor to the moat) that defend it.
Building the future bench
The final problem is formation. If clients no longer fund junior training through billable hours, firms must make an intentional investment in simulators, shadowing and apprenticeships — or face a talent-bench shortfall like the ones documented in Stall Points. However, incentives get in the way: a partner retiring in three years has little reason to fund a bench that pays off in ten. Après moi le déluge. The firms that resist that logic, Usman argues, are the ones that will still have signers in 2031.
About the Guest
Usman Sheikh is the founder and managing director of High Output Ventures, a venture studio building and backing technology-augmented service firms, and the author of Framebreak, a weekly newsletter on strategy, consulting and the knowledge-work economy. Connect with him on LinkedIn.
About the Host
Aidan McCullen is the 2025 Thinkers50 Innovation Award recipient, a keynote speaker on AI, disruption, innovation and change, host of The Innovation Show, and author of Undisruptable: A Mindset of Permanent Reinvention (Wiley). Learn more at theinnovationshow.io/about-aidan-mccullen.
About The Innovation Show
The Innovation Show is the Thinkers50-recognised podcast where square pegs find their place in a world of round holes. Each week, Aidan McCullen hosts world-class authors, scientists and practitioners on disruption, innovation, change, transformation, leadership and creativity. This series is brought to you by Kyndryl, who run and reimagine the technology systems that drive advantage for the world’s leading businesses — learn more about Kyndryl and the Kyndryl Institute at kyndryl.com. Subscribe to the Thursday Thought on Substack for weekly essays on innovation and change.
Listen or watch now: Apple Podcasts | Spotify | YouTube
Frequently Asked Questions
What is the intelligence core?
The intelligence core is Usman Sheikh’s term for a feedback loop in which AI produces work, human experts error-correct it, real-world results flow back, and every correction is fed into the system so the next output improves. It replaces headcount as a firm’s source of leverage and can shift a firm’s structure from a pyramid to a ring.
Why did consulting firms adopt the pyramid structure?
The pyramid solved a scarcity problem: senior judgment was the firm’s most valuable asset, but partners had limited hours. Layers of juniors scaled that judgment, generated margin through the gap between billing rates and salaries, and served as a client-funded training ground for the next generation of leaders.
How is AI different from offshoring or RPA?
Offshoring changed where work got done and RPA changed how predictable processes ran, but both left the pyramid intact. AI changes who does the work by producing “judgment-shaped artifacts” — outputs that look like expert judgment but contain none — which removes the entry-level work juniors once learnt on.
What does “fluent but substitutable” mean?
It describes professionals who can operate an LLM fluently but cannot grade its output. Because they add no judgment the machine lacks, anyone with the same prompt — or eventually the machine itself — can replace them. Usman Sheikh’s advice is to do the work by hand until you can reliably assess what the machine produces.
What is the AI discount?
The AI discount is clients demanding lower fees because they assume their advisers now use AI to do the work faster — KPMG reportedly asked its own auditor for one. It is one of two main ways the AI surplus drains away from firms; the other is rent extracted by the AI platforms firms build on.
How do firms train junior talent when AI does the entry-level work?
Firms must fund formation deliberately rather than relying on client-billed hours — through simulators, structured shadowing, apprenticeships, and exercises where juniors grade AI output before seeing the senior’s answer. The obstacle is incentives: leaders near retirement have little reason to invest in a bench that pays off years later.