In January 2026, McKinsey's Global Managing Partner Bob Sternfels told the HBR IdeaCast that his firm's headcount was 60,000: 40,000 humans and 20,000 agents. The firm later revised the agent figure to 25,000. That one statistic captures how far strategy consulting has travelled in two years, from experimenting with chatbots to staffing agents alongside consultants as standard equipment. This article sets out what the major firms have actually deployed, what it has measurably saved, and why the way firms judge agents is changing.
Key takeaways
- An agent takes an objective and runs a workflow to reach it, calling external tools along the way; a chatbot only answers prompts. The major firms now give consultants agents by default.
- McKinsey reports 25,000 agents in a 60,000-strong workforce and aims for every employee to be enabled by at least one agent within about 18 months. BCG consultants have built more than 18,000 custom GPTs. Bain chose to buy (Sage with OpenAI, ChatGPT, Claude) rather than build.
- The measured payoff is time: at BCG, about 15% less time on low-value work, with roughly 70% of the saved hours reinvested in higher-value analysis.
- The evaluation question is moving from how many agents exist to how much value they create, and humans remain accountable for the output.
Chatbot or agent: the difference that matters
The distinction is practical, not academic. A chatbot answers the prompt it is given. An agent takes an objective, runs a multi-step workflow to reach it, and can call external tools along the way: search a knowledge base, query a dataset, draft a document, then check it. For a consulting firm the difference is the difference between a consultant asking a question and a consultant delegating a task. It is also why agents raise the stakes on verification: the more steps a system executes on its own, the further an early error can travel before a human sees it.
McKinsey: Lilli, then 25,000 agents
McKinsey is the clearest case of a firm building its own stack. Its push started early: in August 2023 it rolled out Lilli, a generative-AI platform that searches and synthesises McKinsey's internal knowledge across more than 100,000 documents. Its AI work runs through QuantumBlack, which has more than 1,700 staff.
“Little over a year and a half ago, that [the number of agents deployed by McKinsey] was 3,000 agents and I originally thought it was going to take us to 2030 to get to one agent per human. I think we're going to be there in 18 months and we'll have every employee enabled by at least one or more agents.”[1]
Workforce of 60,000 including 25,000 AI agents (the figure McKinsey calls most accurate, revised up from the 20,000 first cited). Target: every employee enabled by at least one agent within roughly 18 months. Lilli launched August 2023 over more than 100,000 internal documents. QuantumBlack: more than 1,700 staff.[2]
BCG: Deckster, GENE and 18,000 custom GPTs
BCG has gone broad rather than deep on a single platform. It gives consultants Deckster for drafting slides and GENE, a conversational assistant the firm uses for presentations, podcasts and outreach. After rolling out ChatGPT Enterprise firm-wide, BCG consultants built more than 18,000 custom GPTs to automate research, slide production and routine queries. That number is worth pausing on: it describes thousands of consultants building their own small tools, not a central team shipping a product.
Consultants now spend about 15% less time on low-value work such as building slides, and reinvest roughly 70% of those saved hours into higher-value work such as deeper analysis. BCG X, the unit that builds custom AI systems for clients, has more than 3,000 employees.[3]
Bain: buying rather than building
Bain took a different route and is the clearest example of a major firm choosing to buy. Sage, built with OpenAI, lets consultants generate insights from Bain's own intellectual property. The firm also runs a customised version of ChatGPT and has more recently deployed Claude. Technology- and AI-enabled work was about 30% of Bain's business in 2025, with more than 2,500 AI projects delivered to date.[4]
Build versus buy is less a technology decision than a statement about where a firm thinks its edge lies. McKinsey is betting that proprietary knowledge plumbing is the moat. Bain is betting that the models are a commodity and the moat is the IP you feed them and the judgment of the people using them. Both can be right; they simply imply different hiring, different governance, and a different exposure to the labs that supply the models.
Beyond the big firms: agents for boutiques
The shift reaches past the large firms. Accessible tools such as Anthropic's Claude have brought agents within reach of specialised boutiques working on tighter budgets. Queen's Tower Advisory runs an 80/20 model, 80% agents and 20% humans, so it scales on agent capacity rather than headcount. Agents are also becoming a product: firms sell them to clients, and a share of what gets built starts bottom-up, when a consultant builds a tool for one client's problem and adds it to the firm repository for the next team that needs it.
Side by side: how the firms approach agents
| Firm | Approach | Named tools and platforms | Scale signals, as reported | Stated philosophy |
|---|---|---|---|---|
| McKinsey | Build | Lilli (Aug 2023, 100,000+ documents); QuantumBlack (1,700+ staff) | 25,000 agents in a 60,000 workforce; one agent per employee targeted within ~18 months | Agents as part of headcount, updated monthly (Sternfels) |
| BCG | Build broadly on a bought foundation | Deckster (slides), GENE (assistant), ChatGPT Enterprise; BCG X (3,000+ staff) | 18,000+ custom GPTs built by consultants; ~15% less time on low-value work, ~70% of it reinvested | Redesign workflows so AI shows up in the P&L, not just token use (Schweizer) |
| Bain | Buy | Sage (with OpenAI), customised ChatGPT, Claude | ~30% of 2025 business tech- and AI-enabled; 2,500+ AI projects | Buy the models, apply them to proprietary IP |
| PwC | Deploy with human accountability | Firm-wide AI training | 315,000+ staff trained in AI | “The human is still accountable”; measure agents by how well people use them (Priest) |
| AI-native boutiques (e.g. Queen's Tower Advisory) | Agent-first | Claude and similar accessible tools | ~80% agents, ~20% humans | Scale on agent capacity rather than headcount |
From how many agents to how much value
The way firms judge agents is changing. Across both corporates and consulting firms, the question is shifting from how many agents exist, or how many people use them, to how much value they create. PwC's Chief AI Officer Dan Priest makes the point bluntly: people, not agents, still run the workforce, and an agent is best measured by how well people use it rather than by how much it could automate in theory.
“The human is still accountable. The humans are the ones who get certified. The humans are the ones who get licensed. The humans are the ones who get empowered.”[5]
What this means for consulting firms
Agent count is a vanity metric; reinvested hours are the real one. BCG's 15% and 70% figures are more informative than any headcount of agents, because they show where the freed time went. A firm that saves time and simply bills fewer hours has automated its own revenue away.
Accountability does not delegate. Every firm quoted here keeps the human responsible for the output. That is not caution for its own sake: the hallucinated-citation incidents of 2025 and 2026 show what happens when a confident system meets an unverified publication. Agents make verification more urgent, not less.
The skill gap moves up the pyramid. When agents do the production, the scarce skill is directing them: framing the task, judging the output, knowing what to delegate and what to keep. That is a training problem before it is a technology problem, and it is the one Spaik's programmes are built around.
This analysis expands chapter 4[6] of The State of AI in Consulting 2026, which also covers the new competitors, the reshaping of the pyramid, revenue and pricing, hallucinated citations, and the EU AI Act.
Notes and sources
- HBR IdeaCast, interview with Bob Sternfels, Global Managing Partner, McKinsey & Company, January 2026
- McKinsey & Company statements on QuantumBlack, Lilli (August 2023) and agent deployment, 2023 to 2026
- Boston Consulting Group statements on Deckster, GENE, ChatGPT Enterprise rollout and custom GPTs
- Bain & Company statements on Sage (built with OpenAI), ChatGPT deployment and Claude adoption
- PwC, remarks by Chief AI Officer Dan Priest on human accountability for AI agents
- Spaik, The State of AI in Consulting 2026, chapter 4
Figures attributed to third parties are their own reported data; Spaik did not produce those statistics. Where the text offers an interpretation, it is Spaik's own.
Continue reading
- The State of AI in Consulting 2026The full report this analysis is drawn fromRead
- AI hallucinations in consultingThe verification problem agents make more urgentRead
- EU AI Act for consulting firmsWhen deploying agents makes you a providerRead
- AI for consulting firmsHow Spaik works with consultanciesRead
Working with Spaik
Agents only pay off if your consultants can direct them
Spaik trains consulting teams to work with AI on real engagement tasks, and advises firms on where agents belong in the delivery model. Hands-on, tool-agnostic, and built around the judgment that keeps humans accountable.