McKinsey says it now has 20,000 AI agents working alongside 40,000 people a year and a half earlier it had about 3,000 (yes, agents counted like staff)

that's CEO Bob Sternfels talking to Harvard Business Review, as reported by HRD.

and it's the clearest sign yet that AI in consulting is changing the shape of the firm, not just the speed of the slides.

if you're aiming for a junior consultant or analyst role, this is the shift to understand before your first interview.

How AI is changing the consulting pyramid

the classic consulting model is a pyramid: a few partners at the top, lots of juniors at the bottom doing research, data cleaning and slide drafts.

AI eats exactly that bottom layer first.

what's happened over the last year:

  1. the agents are scaling. Sternfels said McKinsey wants to "get to one agent per human" over the next 18 months, per The Irish Times
  2. starting pay is frozen. McKinsey and BCG held 2026 graduate offers at 2025 levels, the third year running of flat starting salaries at top firms. the Big Four haven't raised starting pay since 2022 (The Irish Times)
  3. intakes shrank. KPMG's UK graduate and school leaver intake fell from 1,399 in 2023 to 942, and Deloitte's from 1,700 to 1,400 (HR Review)
  4. the first year got rewritten. this month all four of the UK's biggest firms told City AM they've overhauled, or are overhauling, their graduate programmes because AI now handles routine first-year work like data entry and basic report drafting (via Resultsense)

but it isn't a straight line down. McKinsey's North American chair said the firm would hire 12% more graduates in 2026 than in 2025.

so the pyramid isn't vanishing... it's changing shape.

What a junior consultant actually does now

KPMG UK's chief people officer, Marc Burrows, told City AM that graduates will become more involved in "reviewing and exercising judgement rather than simply producing outputs".

so the job moves like this:

→ before: find the data, build the model, format the deck → now: check the AI's data, challenge the model, decide what the deck should say

KPMG has even moved professional exams to the start of its scheme so graduates can advise clients from their second year.

faster ramp, less room to hide.

What uni won't teach you about AI in consulting

case interview prep still trains you to do every bit of the analysis yourself, under time pressure. useful, but it's now only half the test.

McKinsey has been piloting interviews where business school candidates use its AI assistant, Lilli, to analyse a case.

assessors look at how candidates prompt it and whether they show "curiosity and judgment" with what it gives back (The Irish Times).

and most degrees don't teach the review skill at all: spotting where a confident answer is quietly wrong.

The critical-thinking move: 4 checks before you trust an AI analysis

use these on any AI output, in a case interview, an internship or your first project:

  1. trace one number to its source. take the most important figure in the output and find where it came from. if you can't, it doesn't go in the deck
  2. ask what the client would push back on. AI gives you the average answer. clients pay for the answer that fits their situation
  3. name the assumption. every recommendation rests on one or two assumptions (growth, cost, timing). write them down and test what happens if they're wrong
  4. separate synthesis from judgement. AI is great at "here's what the data says". the junior's value is "here's what we should do, and why"

What to do this month

  1. practise two case interviews with an AI assistant allowed, then have a friend grade how you questioned its output, not just your final answer
  2. pick a public annual report, ask an AI to summarise it, then fact-check five of its claims against the original
  3. get properly good at one AI-assisted tool consultants use every day, like Copilot in Excel, so you can explain any model you build line by line
  4. read how firms describe their graduate schemes this year and look for the words "judgement" and "review". that's what they're actually hiring for

so TLDR: AI isn't deleting the junior consultant, it's deleting the junior's old job. the new one starts at judgement.

which part of a junior consultant's job do you think should never be handed to an agent?

Sources

Get the next issue by email

theVANTAGEd is written by students to bring critical thinking to AI across real industries. One issue every two weeks.