How McKinsey Plans to Survive AI (and Reinvent Consulting)

How does a storied consulting firm reflect on its history while forging a path ahead in uncertain times? In this episode of HBR IdeaCast, McKinsey Global Managing Partner Bob Sternfels speaks with host Adi Ignatius about the controversies that have ignited change at the large consulting firm, how exactly they are reorganizing human talent in an age of AI, and what business models he thinks will be most successful as the industry shifts. — channel description

A 31:36 episode of HBR’s IdeaCast podcast (also published as a YouTube video on the HBR channel) in which Bob Sternfels, McKinsey’s Global Managing Partner since 2021, reflects on the firm’s 100-year anniversary and articulates the consulting business model McKinsey is migrating toward. Hosted by Adi Ignatius (HBR editor-in-chief). Manual (human-curated) caption track — the wiki’s first non-ASR video source.

This is the wiki’s clearest first-party McKinsey source to date. Prior McKinsey content has come through wiki-adjacent channels: the Rewired 2nd-edition sample (Lamarre / Smaje / Levin / Singla / Sukharevsky / QuantumBlack 2026) and the McKinsey Global Survey numbers folded into AI Index reports. Sternfels here speaks for the firm itself — its self-narrative around AI, its post-controversy compliance overhaul, its hiring-process re-tooling, and its bet that the industry shifts from advisory / fee-for-service to outcome-underwriting.

TL;DR

• The headline number: McKinsey is running a 60,000-strong workforce composed of 40,000 humans + 20,000 agents. Eighteen months ago that was 3,000 agents. Sternfels originally projected reaching one agent per human by 2030; he now expects to reach that ratio in 18 months (~mid-2027). “We’re going to have a workforce that is human and agentic.”

• The business-model claim: McKinsey is migrating away from pure advisory / fee-for-service toward an outcome-based model where the firm underwrites the business case alongside the client (“We collectively signed up for this outcome together, and we’re tied on this journey all the way through until that impact is delivered”). About one-third of revenues today are outcome-underwritten; Sternfels’ aspiration is majority-of-revenues by the time he finishes his term. This is the wiki’s strongest single articulation of the consulting industry’s version of the broader Nishar-Nohria outcome-pricing trend.

• The consulting-economics diagnosis: “Half, if not more, of the secret sauce is organizational change as opposed to technology implementation.” Direct corroboration of the micro-productivity trap from a different vantage — McKinsey is naming the same gap (high tech investment, low enterprise-level value) Bain/OpenAI named in Dutt-Chatterji 2026.

• The CEO’s classic dilemma that Sternfels reports hearing in “the truth room”:

  “Bob, do I listen to my CFO or my CIO? My CFO is in my ear that we’re spending a lot of money on technology, but we’re not yet seeing enterprise-level value. … CIO’s saying: are you crazy? This is one of those moments. And if we’re not in the lead, we’re going to get disrupted.”

• The organizational-redesign mechanism: AI enables flatter organizations that cut middle layers and workflow consolidation across departmental boundaries (Sternfels’ worked example: a mortgage process that runs through origination → credit scoring → collection → after-service today as four or five departments could collapse to a single AI-enabled flow). “Why do you have four or five departments in a process if you could really enable this through AI?”

• Four durable leadership skills the models don’t have (Sternfels’ explicit list, in his exploration mode):

  1. Aspiration-setting — “They’re not good at setting the right aspiration level. … Great leaders help an organization set the right ask: what should we aspire to?”
  2. Judgment — “There’s not truth in the model. There’s not judgment in the model. Humans need to impose those parameters.”
  3. Discontinuous-leap thinking — “They’re great at a linear approach to problem solving. … What they’re not great at is discontinuous leaps, truly novel thinking. So we’re going back to liberal-arts degrees and saying, hey, let’s come back to some of the things that might have been deprioritized in the past.”
  4. Human-to-human skill (implicit; surfaces via the hiring section) — “If you’ve done a team sport or if you’ve worked in retail working your way through college, you’ve had to engage with others. It’s built a skill of human-to-human skill that we’re now indexing a lot more on.”

  This list converges on the collaboration / creativity / critical-thinking framework of Globerson et al. 2026 from a hiring-criteria angle.

• Hiring overhaul via self-applied analytics: McKinsey applied “analytics on ourselves” — 20 years of data, what predicts making partner. Three biggest biases the analysis surfaced (out of 50 implications total):

  1. Setback-and-recovery beats perfect marks for partner-track resilience.
  2. Team-sport / retail-job experience (i.e. real human-to-human work) was under-weighted.
  3. Aptitude-to-learn-new-stuff beats subject-mastery of the chosen major. “We purposely create an environment where no human in the world will have any pattern recognition. … We figure out how well do you do in an environment where you have no pattern recognition, you just have to go figure things out.”

  Result: McKinsey changed the assessment process to test for resilience, collaboration, and aptitude-in-novel-environments. Indirectly substantiates durable-skills from the recruitment/selection side — McKinsey isn’t measuring durable skills of workers, it’s measuring them of candidates.

• Post-controversy governance overhaul ($1B invested over ~6 years):

  • Imported leadership from outside professional services: “I brought in the head of internal audit from Apple, the head of compliance from Walmart” to modernize processes.
  • Adopted publicly-traded-company-equivalent compliance and accountability standards despite remaining privately held.
  • Clarified the partnership social contract: “You [partners] all grew up with the idea that one of the privileges of a partner is to commit the firm. I’m not saying that’s not the case anymore, but you don’t do it alone, you do it with risk professionals.”
  • The named controversies: opioids / OxyContin work and work in South Africa (state-capture / Eskom-Gupta era — referenced as the firm’s “partnerships in South Africa”); plus conflict-of-interest accusations in the U.S. and elsewhere.
  • Apology framing: “We apologize. We got those wrong. … We don’t want to set out just to remediate the problem; we want to set out to try and set the standard for professionalism for our industry.”

• Push-back on the climate-transition criticism: Sternfels distinguishes humble-camp learnings (opioids, South Africa) from courageous-disagreement-camp positions. Example of the latter: McKinsey’s work with hard-to-abate sectors. “It was the old McKinsey’s accelerating climate degradation, etc. And we pushed back and said: no, look, if you’re going to be committed to climate transition, you have to work with the hardest-to-abate sectors. It’s just naive to say that you’re going to solve this problem without that.”

• Three CEO-level paradigm shifts Sternfels reports hearing transversally (across geographies and industries):

  1. “How do I get value from this technology?” (the AI-transformation question).
  2. “How do I build more institutional resilience?” — “unfortunately things will never go back to how they were. There’s going to be a world of continuous shocks. … I need to play offense and defense at the same time.”
  3. “What should my future organization model be?” — Sternfels references a 1959 HBR paper by his predecessor (referred to as “Gil Klee”; ASR-uncertain proper-noun rendering) as the precursor to the matrix organization, now under universal pressure.

• Three things great leaders most consistently get right (Sternfels’ framing of where they go wrong):

  1. “Hunger and thirst to acquire new information … a ruthless quest for let me continue to question new things.”
  2. Collaboration across the value chain — “We’re seeing when people collaborate across the value chain find disproportionate gain. And yet our organizations aren’t geared to collaborate well.”
  3. Speed — “Faster organizations outperform slower organizations, even if they make more mistakes. And yet we’re not wired to do that. There’s such risk aversion in large enterprise.”

What was actually ingested

Full ~31:36 video transcript via the manual (human-curated) English caption track — kind: manual per the YouTube caption-track metadata. Not ASR. Speaker labels (BOB STERNFELS / ADI IGNATIUS) preserved verbatim from the manual track; light line-break normalization and ~12-second-window grouping with speaker-turn breaks (every new speaker forces a new line). All 12 microformat chapters covered end-to-end.

Per §Source-page conventions specific to videos, author: carries the channel name (Harvard Business Review). Speaker names — Bob Sternfels (guest) and Adi Ignatius (host) — are named in body prose. Both are first-mention by name in this source; per the author-entity-promotion rule, neither is promoted to entity page on a single source.

Detailed walkthrough

McKinsey at 100 — co-creation as the firm’s self-narrative

Adi Ignatius opens with a positioning question: “how would you summarize the company’s 100-year legacy? To what extent has McKinsey created the ideas that have shaped the business world or to what extent is it about identifying and suggesting best practices that come from elsewhere?”

Sternfels’ framing rejects the dichotomy: McKinsey co-creates with clients to help them get to places they can’t get to themselves. He estimates the split as roughly half novel co-creation with clients, half importing innovation across geographies and sectors to clients who lack the access. The firm invests over $1 billion a year in innovation — McKinsey Global Institute (research arm) and McKinsey Health Institute named as institutional anchors of the proprietary-IP claim.

AI’s impact on clients (vs. on McKinsey itself)

Sternfels splits the question into two coins:

Client-side. Two themes he hears across geographies and industries:

• Enormous belief in the technological wave — productivity gains in customer care, back-office processes; growth potential in shortened drug-discovery timelines, etc.
• The CFO-CIO truth-room dilemma — CFOs reporting AI spend without enterprise-level value; CIOs warning that being a fast-follower is no longer safe. The named resolution: “What we’re finding is half, if not more, of the secret sauce is organizational change as opposed to technology implementation.”

McKinsey-side. Two structural shifts:

• Workforce composition — 60,000 ~= 40,000 humans + 20,000 agents (up from 3,000 agents 18 months prior); on track for 1:1 human-to-agent ratio in ~18 months.
• Business model — pure advisory → outcome-underwriting (currently ~33% of revenues; aspirationally majority by end-of-Sternfels-term).

Why clients still pay McKinsey if AI commoditizes analysis

Adi presses: “if technology can continue to commoditize even the kinds of analysis and insight that a McKinsey has long provided, what will clients actually be paying for?” Sternfels’ answer is structural rather than defensive:

“The kind of problems that we have tackled with our clients over 100 years has not been static. … I’m now considered a dinosaur in our firm because I’m a little over 30 years with us. But the stuff that I did when I joined as an associate 32 years ago, we wouldn’t consider even doing right now. Why? Because clients do that stuff themselves. … What this is going to then mean is just going to be that next evolution of — there’ll be a whole bunch of things that a couple of years ago we did for our clients that our clients will do for themselves. And the imperative will then be to move to the even more complicated questions. … They’re going to pay us to find ways to double their market cap. And until we get to CEOs who say I don’t want to double my market cap, I think they’ll always be a more complicated set of questions and opportunities out there.”

This is a longitudinal version of the AI deskilling task-composition shift — McKinsey describes 30 years of consultant work-mix migrating upward as lower-tier tasks commoditize. It’s also a strong restatement of the Anthropic Economic Index 5’s skill-biased technological change framing: the demand for higher-tier task work doesn’t go away when lower-tier tasks automate, it intensifies.

The hiring overhaul — analytics on themselves

Sternfels narrates the discovery: when he became GMP four years prior, the talent team reported “a million applications a year, hire 8,000 to 10,000 people, brightest minds applying.” Sternfels kept asking “how many profiles are we really looking for? What are we systematically screening out?” and discovered only 500 pathways worldwide led to McKinsey hires.

McKinsey applied analytics on 20 years of internal data: “What are the skills and characteristics that are most likely to make partner in McKinsey? Because it’s not perfect, but it’s a marker of success. Only 1 in 6 hires make partner.” They surfaced 50 implications; the three named ones (paraphrased earlier in the TL;DR) all involve relaxing rigid credential-based filters in favor of behavioral/aptitudinal ones. The post-overhaul assessment now includes a deliberate no-pattern-recognition environment — testing how candidates perform when they can’t fall back on memorized patterns.

Post-controversy governance — the humble-vs-courageous distinction

Adi raises the unwelcome publicity: “OxyContin, bribery charges in South Africa, conflict-of-interest accusations in the US and elsewhere. How do you account for all of that?”

Sternfels’ framing distinguishes:

Humble-camp — opioids, South Africa partnerships:

• “What we learned is that we have to have a higher diligence around client selection.”
• New framework assesses country / topic / institution / individuals / operating environment per engagement.
• “You [partners] grew up with the idea that one of the privileges of a partner is to commit the firm. … You don’t do it alone, you do it with risk professionals.”
• “$1 billion invested. Brought in the head of internal audit from Apple, the head of compliance from Walmart to basically modernize us around these processes.”
• Same compliance and accountability standards as a publicly-traded company, despite being private.

Courageous-disagreement camp — work with hard-to-abate sectors:

• “It was the old McKinsey’s accelerating climate degradation, etc. And we pushed back and said no, look, if you’re going to be committed to climate transition, you have to work with the hardest-to-abate sectors.”
• The argumentative thicker-skin posture is named explicitly: “There’s a portion of this that was rooted in just more transparency, more criticism. And I’m trying to build a bit of thicker skin.”

The compliance overhaul includes a constraint Sternfels names directly: “Growth has never been our objective function. … The ethos internally is that we’re a profession, not a business, meaning that we want to put our clients’ interests ahead of ourselves. … But without the right controls, you can’t guarantee that. … Even though we had that as our ethos, if you don’t put in place compliance, you’re not going to actually be able to enforce those standards. And it’s painful for a partnership to give up some of that autonomy.”

From PowerPoint strategy to outcome-underwriting

Adi’s challenge: “people would say consultants prescribe frameworks but don’t have to live with the consequences. How do you ensure that McKinsey’s advice is more than a PowerPoint strategy?”

Sternfels’ counter:

“I hope we get out of PowerPoint entirely some day — we’d love that, Microsoft — but it isn’t about providing a unique insight. … What we really aspire to be is to be impact partners with our clients, and we’re on a change journey of moving, quite frankly, from a model that was advisory to one that underwrites outcomes. And today, about a third of our revenues total are underwriting outcomes. So it’s not ‘hey, hand me a PowerPoint, great’ — we collectively signed up for this outcome together, and we’re tied on this journey all the way through until that impact is delivered.”

This is a sharp consulting-side restatement of Nishar-Nohria 2026’s outcome-based pricing as a named industry signal. Where Nishar-Nohria observed it at SaaS vendors (Adobe), Sternfels describes McKinsey’s own internal migration. The economic mechanism is identical: when AI commoditizes the act of advice/output, value migrates to who’s accountable for the result.

Continuous shocks and the matrix-organization question

Sternfels’ three-theme framing of CEO-level questions surfaces a 1959 HBR paper by “Gil Klee” (ASR-uncertain rendering of the name; likely Gil Clee — Marvin Bower-era McKinsey, predecessor as Managing Director, the GMP equivalent) as the precursor of the matrix-organization concept. Today, Sternfels says, “if you look at almost every large enterprise today, there is some version of a matrixed organization. And I hear different tension points from CEOs about why their org is one of the bottlenecks … Whatever it may be, I hear a lot of questions about what should my future organization model be.” The flatter-org-via-AI prescription from earlier in the conversation reads as McKinsey’s current best answer.

Closing — the leadership-factory legacy

Sternfels names what he wants McKinsey to remain known for in 10 years:

• Continuing: “leadership factory of the world. … We produce more CEOs than any other institution in the world.”
• New: completing the journey from advisor to impact partner. “It’s not ‘they gave me great advice, and if it worked it was because they were smart, and if it didn’t work it’s because I didn’t implement’ — that’s the joke. It’s: we designed a business case together. These guys underwrote the same outcomes that I took to the board, and we went on this journey, and we kept at it until we got someplace I didn’t think I could get to.”

Cross-source resonance

Source	Connection
Lamarre, Smaje, Levin et al. (Rewired 2nd ed)	The McKinsey institutional view from a different layer. Lamarre et al. give the practitioner playbook (6 capabilities, 70% talent-density shifts, 20% EBITDA / $3:$1 ROI, 1–2-yr breakeven); Sternfels gives the firm-as-vendor self-narrative (40k humans + 20k agents, outcome-underwriting, post-controversy governance, hiring-process overhaul). The two sources should be read together for the McKinsey organizational thesis.
Dutt-Chatterji 2026 (Bain, OpenAI)	Sternfels’ “half-or-more of the secret sauce is organizational change” is exactly the trap diagnosis Bain/OpenAI named, from a McKinsey vantage. Two consulting firms converging on the same diagnosis is structural evidence that the trap is real — not just a Bain framing.
Nishar-Nohria 2026	The outcome-based pricing trend is named here as Adobe; Sternfels is the consulting-side worked example. Same mechanism (AI commoditizes the advisor function → value migrates to outcome-accountability) at two different industries.
Anthropic Economic Index 5	Sternfels’ “stuff I did 32 years ago we wouldn’t consider doing right now — clients do that themselves” is a 30-year longitudinal restatement of the AEI 5 task-composition shift. Direction-of-travel matches.
Globerson et al. 2026	The four durable leadership skills models lack (aspiration / judgment / discontinuous thinking / human-to-human) is McKinsey’s variant of the collaboration / creativity / critical-thinking framework, articulated as hiring criteria rather than as assessment instruments. Direct convergence.
Anand-Wu 2025	Anand-Wu’s 2×2 (cost of errors × type of knowledge) implies the high-cost-of-errors / explicit-knowledge cells are exactly where consulting firms can underwrite outcomes credibly. McKinsey’s outcome-underwriting move can be read as concentrating revenue in the cells where the firm can afford to be on the hook.
Warner & Wäger 2018	Sternfels’ “world of continuous shocks … offense and defense at the same time” is a folk-version of the dynamic-capabilities sense/seize/transform framework. The Teece tradition would frame Sternfels’ three CEO themes as the ongoing strategic-renewal conditions.
Werner & Le-Brun 2025	Sternfels’ “flatter-org-cuts-middle-layers” prescription is operationally compatible with the Octopus Organization frame — distributed cognition, less coordination drag, more local adaptation. McKinsey is signaling it sees the same trajectory the AWS executives in residence prescribed.
durable-skills	Sternfels names four named durable leadership skills models lack. Concept-page minor extension warranted.
micro-productivity-trap	Sternfels’ direct corroboration of “tech spend high, enterprise value low, organizational change is the missing variable.” Concept-page minor extension warranted.
ai-deskilling / automation-vs-augmentation	Sternfels’ 32-year longitudinal account of consultant work-mix migration is extra evidence for the task-composition-shift mechanism.
enterprise-ai-adoption	The 60K=40K+20K workforce datapoint and the 18-month forecast for one-agent-per-human are concrete adoption-pace anchors at the consulting-firm level. Concept-page entry warranted.

Linked entities and concepts

• Channel (1st mention by name as a video-source author, but already an entity): Harvard Business Review — already in the wiki. Add this source to its inbound set.
• Host: Adi Ignatius — editor-in-chief of HBR; long-time IdeaCast host. Originally Dangling on this source’s first ingest (per author-entity-promotion rule); promoted to entity page on 11 May 2026 following his second wiki-source appearance as interviewer on Jassy 2025.
• Guest (1st mention by name; Dangling): Bob Sternfels — Global Managing Partner of McKinsey & Company since 2021. With the firm for 32+ years. Body-prose mention only; not promoted on a single source.
• Organizations:
  • McKinsey & Company — already in the wiki; substantively extended by this source. The wiki’s first-party Sternfels source. Add this source to McKinsey’s inbound set.
  • Harvard Business Review — publisher / channel.
  • McKinsey Global Institute — McKinsey’s research arm; already mentioned in passing across the wiki.
  • McKinsey Health Institute — first wiki mention; defer entity promotion (single source).
  • Apple, Walmart — referenced as the source of compliance / internal-audit hires; not promoted on a single mention.
  • Microsoft — referenced as the home of PowerPoint, with a wry “we’d love to get out of it” aside.
• People referenced in passing:
  • Gil Clee (likely; ASR-uncertain) — McKinsey Managing Director / GMP-predecessor; author of a 1959 HBR paper on creating a global organization, named as a precursor to the matrix-organization concept. First wiki mention; defer entity promotion (single source).
• Concepts touched (most are minor extensions to existing pages):
  • durable-skills — Sternfels’ four-skill list (aspiration / judgment / discontinuous thinking / human-to-human). Minor concept-page extension.
  • micro-productivity-trap — direct corroboration. Minor concept-page extension.
  • ai-deskilling — 32-year longitudinal consultant-work-mix narrative. Minor concept-page extension.
  • enterprise-ai-adoption — McKinsey’s 60K-with-20K-agents workforce datapoint. Minor concept-page extension.
  • automation-vs-augmentation — Sternfels frames it as human-and-agentic workforce rather than human-vs-agent. Minor concept-page mention.
  • ai-employment-effects — Sternfels’ implicit claim is that consulting headcount stays / grows but composition shifts. Minor concept-page mention.
• Candidate concepts (single-source for now; defer promotion):
  • Outcome-based services / outcome-underwriting — Sternfels names this multiple times; Nishar-Nohria supplies the cross-industry frame. Two-source convergence threshold reached in this batch — promotion candidate for a follow-up sweep.
  • Agent-augmented workforce ratio (humans:agents) — Sternfels’ 60K=40K+20K is the wiki’s first concrete vendor-level number; promote on second-source mention.
  • Continuous-shocks resilience — Sternfels’ offense-and-defense framing.
  • Post-controversy governance overhaul (consulting-firm subtype) — single-source for now.

Notes on confidence and lifecycle

• Source quality: a long-form interview on a flagship management-publishing platform (HBR IdeaCast, ~145K views), with a manual (human-curated) caption track. Authoritative on McKinsey’s self-narrative about its AI strategy, hiring, and post-controversy governance. Not externally validated; figures Sternfels cites (40k humans + 20k agents; one-third of revenues outcome-underwritten; $1B compliance investment; 1-in-6 hires make partner; 20-year self-applied analytics) are firm-self-reported and not cross-checked here. Treat the direction-of-travel as well-substantiated, the specific magnitudes as best-available-self-report.
• Page does not carry confidence: — sources are evidence, not claims (per §Lifecycle).
• Transcript provenance: Manual (human-curated) caption track is materially higher quality than the four prior kind: asr video sources. Speaker labels survive intact; proper-noun reconstruction is unnecessary.
• Ambiguous proper noun: “Gil Klee” — the manual transcript gives this rendering; the historical figure is almost certainly Gil Clee, McKinsey Managing Director from 1967–1973 (Marvin Bower’s successor), known for early HBR pieces on global organization design. The 1959 dating Sternfels gives is approximately consistent with Clee’s mid-career period; the ASR-uncertainty marker [? Gil ?] [? Klee ?] suggests the transcribers themselves were uncertain. Worth flagging if the wiki ever creates a Gil Clee entity page.
• Date positioning: published 9 February 2026; ingested 10 May 2026. Three-month latency (publication-to-ingest) — consistent with the wiki’s reactive-ingest discipline.

What this source does not do

• It does not substantiate the 40k/20k workforce composition externally. No employment-records check, no third-party audit of agent count, no breakdown of what counts as “an agent” inside McKinsey. Treat as firm-self-report.
• It does not quantify the quality delta of outcome-underwritten engagements vs. fee-for-service ones. Sternfels asserts “aligned interests”; no benchmark, no internal study cited.
• It does not address how outcome-underwriting prices. Risk-pricing for an underwritten engagement is implicit (the consultancy is taking on tail risk), but no economics are laid out.
• It does not engage with critics’ substantive arguments about specific past engagements (opioid work, South Africa). The framing is what we learned and what we changed; the underlying decisions and their accountability chain are not re-litigated here. Sternfels’ “we apologize” is a positional move, not a structural account of how the decisions were made.
• It does not quote agent / harness vocabulary the wiki has built up around engineering-side AI. Sternfels names agents and agentic workforce but doesn’t engage with harness / agentic-engineering vocabulary at any technical depth — these are operating-side concerns the consulting-firm-CEO frame does not enter.
• It does not cross-validate the durable-skills framing against external scholarship (e.g. Globerson et al. is not cited; nor are any academic frameworks on creativity / discontinuous-leap thinking / aspiration-setting). Sternfels is articulating McKinsey’s intuitions, not synthesizing the literature.
• It does not address how McKinsey hires for AI engineers (vs. consultants) — the QuantumBlack hiring funnel is unaddressed here. The hiring discussion is about the consultant pathway specifically.
• It does not address the quality controls inside the agent workforce. “40k humans + 20k agents” makes the headcount claim; nothing on what the agents do, who supervises them, or how their output is graded.

These are reasonable elisions for a 31-minute interview; flag as gaps a follow-up source (a McKinsey internal report, a third-party study, a Sternfels successor’s account) would need to fill.