I did a full market research exercise with 12 agents and €200.
That number alone says nothing about the time I contributed, the scope of the study, or how it compares with a commissioned consulting engagement.
What should consulting cost if agents reduce the effort needed to deliver research of comparable quality?
For work that agents can perform reliably, the answer should depend more on the person who structures the analysis, challenges its conclusions, and helps the client decide. A smaller practice can acquire research capacity that previously required a larger team. The consultant’s own competence becomes more central when access to that capacity is less distinctive.
That could favor a more personal form of consulting: an adviser whose reasoning you can examine, whose advice you can challenge, and who stays involved when a recommendation meets reality.
The pyramid made sense when analysis consumed people
The familiar consulting team combines senior direction, project management, and junior analytical work. Someone frames the engagement; others collect information, work through it, and prepare material for review.
David Maister’s account of consulting firms explains how this mix shapes their economics. Senior professionals can serve more clients by working with junior staff, while the firm earns a surplus from delivering work with an appropriate mix of skills and costs. The arrangement also provides an apprenticeship. (The Anatomy of a Consulting Firm)
Agents put pressure on the parts of that structure built around repeatable research and production. A consultant can delegate searches, extract facts from documents, compare competitors, and generate alternative analyses without assigning each activity to another person.
McKinsey reports that colleagues using its Lilli platform have saved up to 30 percent of their time searching and synthesizing knowledge. That is the firm’s own report of a bounded activity, rather than a measure of an entire engagement’s productivity. (McKinsey’s Lilli case study)
Incumbents can use agents to compress their own teams and extend proprietary knowledge. A small practice gains an opportunity where its total delivery effort—including framing, coordination, verification, review, and follow-through—falls enough to make a leaner structure viable at the required quality and scope. Agent access alone does not establish that advantage.
When the same quality of preparation takes less labor, staffing becomes a weaker justification for the fee. Clients can reasonably ask what additional expertise, information, or help the engagement supplies.
Twelve agents still need one coherent question
Consider a hypothetical company deciding whether to enter a new market. It could send agents to investigate competitors, prices, customer segments, distribution channels, and regulation. Each might return a substantial report.
The company could still be left with five answers to questions that do not settle its decision.
An attractive market is not necessarily an accessible one. Customers may have a problem but no budget to solve it. A competitor’s public pricing may tell you little about discounts, procurement, or the cost of acquiring customers. The company may lack the distribution relationships that make the opportunity viable.
Start with the choice the company faces: enter now, run a limited test, partner with someone already present, or leave the opportunity alone. The research needs to distinguish between those options.
Define the market consistently. Separate demand from the ability to reach it. Identify the assumptions that would overturn the recommendation. Assign investigations that test those assumptions, including the strongest case against entering.
Anthropic’s research agents duplicated work or misunderstood assignments when their tasks were poorly specified. Its engineering account describes the objectives, output formats, source guidance, and task boundaries used to direct them. Parallel research helped because the work could be separated and its findings brought back together. (Multi-agent research system)
A dozen reports agreeing with one another may simply repeat the same underlying source or assumption. Each investigation should help distinguish the client’s options. Headcount, human or artificial, is a poor substitute for testing the argument.
Sense-making has to leave evidence behind
If a consultant says their value lies in judgment, the client should be able to examine it. “Trust my experience” is an expensive answer when the client can generate a competing analysis before the next meeting.
Explain why this market definition matters, how the evidence was weighed, what an apparent contradiction changes, and what remains unknown. Then show how the client’s capabilities affect an opportunity that might work for somebody else.
For the hypothetical market entry, public research might suggest strong demand while customer conversations reveal that buying requires approval from a department the company cannot yet reach. A useful adviser investigates that conflict and changes the proposal accordingly. A larger slide deck cannot resolve it by repetition.
Agents can help with this work too. They can propose structures, compare explanations, challenge recommendations, and identify missing evidence. The distinction between grunt work and valuable analysis will move as their capabilities improve. A consultant who merely relabels familiar tasks as uniquely human judgment has a fragile proposition.
In an experiment involving 758 BCG consultants, GPT-4 helped participants complete tasks within its capabilities faster. On one problem deliberately selected outside its capability frontier, AI users were 19 percentage points less likely to reach the correct answer. These results concern a particular model and task design, not a fixed limit on today’s agents. (Dell’Acqua and colleagues)
A consultant should show which decisive claims were checked against original evidence, which alternatives were tested, and where the research had to stop. Uncertainty that requires customer interviews, a pilot, or information unavailable on the web belongs in the recommendation.
A claim to superior intuition is harder to distinguish from a confident performance.
The names on the buildings were once people
Arthur D. Little was founded in 1886 by Arthur Dehon Little and Roger B. Griffin, initially working in chemical analysis and industrial research. James O. McKinsey established his firm in 1926, bringing a background in management accounting. (Arthur D. Little’s history, McKinsey’s history)
That history offers a metaphor, rather than proof of a lost golden age. Consulting already includes trusted individual advisers, and large firms still depend on them. Clients also buy institutional continuity, specialist networks, and quality controls that go beyond one person’s reputation.
Agents could reduce the supporting organization a capable adviser needs for a particular engagement. Where total effort falls while quality holds, someone with deep knowledge of a sector might undertake a bounded study with a small practice, supported by agents, rather than building a permanent team for every component of the work.
The client could work directly with the person shaping the analysis. Fewer handoffs could make it easier to challenge an assumption, change direction, and understand how the recommendation was reached.
A name is useful when it identifies responsibility. Celebrity adds little if the famous person sells the engagement and disappears. The relevant consultant is the person who can defend the reasoning, admit a mistake, and help repair it.
A €200 research exercise does not buy a whole firm
Some decisions require proprietary data, access to customers and experts, several specialist disciplines, or sustained implementation across countries and business units. Obtaining reliable evidence can itself be difficult and expensive.
A smaller adviser can also be a bottleneck. Their availability, knowledge gaps, and ability to check their own work matter. A trusted firm may supply independent review, continuity, and delivery capacity that an individual with agents cannot reproduce. Large firms can use the same technology while adding resources that remain scarce.
Where a client mainly needs a bounded analysis of available evidence, the case for a large delivery structure deserves closer examination.
Automation also creates an apprenticeship problem. Junior researchers learn by tracing sources, discovering why numbers disagree, and watching recommendations fail. Firms that delegate much of that work need another deliberate way to develop those skills. Experienced judgment cannot remain the selling point if nobody is learning to acquire it.
Compare the same scope, evidence access, review standard, human effort, and support after the recommendation. My €200 figure cannot settle that comparison by itself.
Buy an analysis you can challenge
For a research-heavy engagement, I would first agree on the decision and what would count as a useful answer. I would ask for a brief covering these points:
| Part of the work | What the client should be able to inspect |
|---|---|
| Decision and scope | The choices the analysis will inform and the boundaries of the market |
| Competing explanations | The strongest case for each option, including doing nothing |
| Evidence and uncertainty | Sources for decisive claims, conflicting findings, and consequential gaps |
| Recommendation | The proposed action, its assumptions, and conditions that would change it |
| Follow-through | A named owner, the next test or action, and a date to review what happened |
For the market-entry example, a short decision memo could recommend a customer pilot, explain why immediate expansion is premature, and identify the finding that would justify investment. Supporting research stays available for inspection. The client gets a conclusion they can use and a way to discover whether it was wrong.
The price still has to reflect the value and responsibility of the engagement. It should also be open to challenge when cheaper production changes the work required. A fixed fee can reward efficiency; a vague promise of results can conceal just as much as a padded estimate of hours.
Twelve agents and €200 make the old relationship between research volume and consulting value worth questioning. The next consulting proposal should make it easy to see whose thinking you are buying, how it was tested, and who will still be there after the presentation.
References
- David Maister, The Anatomy of a Consulting Firm, 2004.
- McKinsey, Rewiring the way McKinsey works with Lilli, company case study, accessed October 3, 2026.
- Anthropic, How we built our multi-agent research system, June 13, 2025.
- Fabrizio Dell’Acqua and colleagues, Navigating the Jagged Technological Frontier, Organization Science, 2026; experiment using GPT-4.
- Arthur D. Little, History timeline, accessed October 3, 2026.
- McKinsey, History of the firm, accessed October 3, 2026.