TL;DR. McKinsey now operates with 20,000 AI agents alongside 40,000 consultants (Sternfels, Guardian 2026) and cut roughly 5,000 roles in 2025 (Fast Company). In the interview, you're no longer asked to walk through a framework: you're asked to challenge Lilli under ambiguity. Get it wrong and you're blacklisted.
Are you prepping McKinsey 2026 with the 2017 edition of Case in Point? Bad news.
The "profitability + 4 buckets" case is dying, replaced by a round where you prompt Lilli live in front of a Senior Engagement Manager.
Do you know what they actually want to see when the AI returns a vague answer?
What actually changed at McKinsey between 2024 and 2026
Three years, three interview logics. And the 2025 → 2026 jump is the most brutal MBB has seen since the PEI arrived.
Before 2024, the format was canonical: a 30-minute case (profitability, M&A, market entry) followed by the Personal Experience Interview. Case in Point logic dominated prep — frameworks drilled, structure announced upfront, hypotheses quantified.
In 2025, two signals tilted the game. First, the layoff wave: McKinsey cut around 5,000 roles over the year (Fast Company). Second, the rise of Lilli, the internal generative AI (launched 2023, scaled 2024-2025), which became consultants' daily copilot.
January 2026, Bob Sternfels appears on HBR IdeaCast and makes the shift official: 20,000 AI agents now work alongside 40,000 humans (The Guardian). A few days later, Delphine Zurkiya (Senior Partner) confirms to Business Insider that interviews now embed Lilli (Business Insider).
For you as a candidate, your 2024 prep is partially obsolete. Not useless — partially obsolete. Critical nuance.
The new McKinsey interview format: prompt, review, judgment
CaseBasix and The Guardian have documented the new round's structure. It runs in three beats.
1. You prompt Lilli. The interviewer poses a business question, you write a structured prompt. Not a "summarize the market" — a prompt that scopes the perimeter, asks for hypotheses, specifies the expected deliverable.
2. You review its output. This is where the trap snaps shut: Lilli deliberately returns vague answers, fuzzy numbers, sometimes half-wrong claims. Stephen Turban (ex-McKinsey, Wall Street Guide) sums it up on BI:
"the AI is created to give information that's not 100% correct or vague — so it's a test of how well students can solve problems with a certain level of ambiguity." (Business Insider)
3. You apply your judgment. You point out what's off, you re-prompt to fill the gaps, you deliver a structured recommendation out loud. The interviewer grades you on three things: your ability to detect the fog, reframe it, and decide without re-prompting forever.
Key difference with Casey at BCG: Casey is a standalone chatbot, neutral BCG-side screening (per Ammon Jensen, BYU MBA, on a DoorDash market-sizing prompt). Lilli, by contrast, is embedded inside a late-stage human round at McKinsey. Your interviewer sees your prompts, your silences, your trade-offs in real time.
What McKinsey no longer tests (and why your classic prep is dragging you down)
Framework drilling isn't dead. It's been demoted. Profitability, M&A, market entry, market sizing: useful for first rounds, insufficient to clear the final.
If you spend 80% of your prep memorizing 12 frameworks, you're misallocating effort. The right ratio in 2026 looks more like 40% frameworks / 60% "AI fluency under pressure".
And there's a concrete risk in ignoring the shift. Marc Cosentino confirms it to Business Insider:
"Some people have already gotten in trouble using AI during case interviews… the interviewers caught on almost immediately, wrapped up the interview, and told the candidates they would not be considered in the future." (Business Insider)
Translation: sneaking ChatGPT into a Zoom case = cross-firm blacklist, instantly. McKinsey, BCG and Bain talk to each other. One sanction at one follows you to the other two.
- ✓Duration: 30 min, fixed format
- ✓Deliverable: recommendation after a MECE framework
- ✓Criteria: structure, quantified hypotheses, mental math
- ✓Tools: paper + pen, no LLM
- ✓Evaluation: candidate alone facing the human interviewer
- ✗Duration: 25-40 min, hybrid human + AI format
- ✗Deliverable: prompt → review → recommendation under ambiguity
- ✗Criteria: judgment under fuzzy output, reframing
- ✗Tools: Lilli provided, no other external LLM allowed
- ✗Evaluation: quality of live human-AI collaboration
The implication for your prep is radical: stop drilling silently on paper, start practicing vocal collaboration with a throttled LLM. It's uncomfortable at first. That's exactly what they're testing.
The 2026 PEI: harder, contextualized for the Lilli era
The Personal Experience Interview stays. The three classic axes (Personal Impact, Entrepreneurial Drive, Leadership) too. But the framing of the stories shifts.
In 2024, a strong leadership story sounded like: "I led a team of 6 on a stalled project, here's how I unblocked it…". In 2026, the same story plays better if you add: "…and the key moment was challenging our prioritization tool's recommendation, which pointed in the wrong direction."
Why? Because the implicit question the interviewer is now asking is: "Describe a situation where you had to arbitrate between a tool's output and your own judgment."
Three PEI reframings to prepare for McKinsey 2026:
- Drive — A moment when you pushed a project even though an analysis (Excel, dashboard, AI) suggested the opposite, and you were right.
- Leadership — A moment when you pulled an overconfident team back from a tool's output to reintroduce productive doubt.
- Personal Impact — A moment when you turned a senior's (or a tool's) vague answer into a concrete action plan.
If you need the PEI fundamentals (structure, timing, common pitfalls), see our MBB interview guide. Here we cover only the 2026 pivot.
EU regulatory frame: GDPR and AI Act when Lilli is in the loop
You're a European candidate being assessed by a US AI deployed on EU applicants. Two texts protect you — and give you ammunition for the back-of-interview questions.
EU AI Act, Articles 5 and 50 (applicable since February 2025). Article 5 bans automated emotion inference in employment contexts. Article 50 mandates transparency: a generative AI system that interacts with you must be identified as such (artificialintelligenceact.eu).
GDPR, Article 22. Any significant decision based solely on automated processing (including profiling) opens a right to human intervention and contestation (CNIL).
And the reputational risk exploded in 2026. The Stack documented an incident in which 46 million chat logs and 728,000 private files were exposed via Lilli (The Stack). The same disclosure points to 266,000+ OpenAI vector stores feeding the retrieval pipeline, and 1.1 million files flowing through external APIs. It's a legitimate question to raise at the end of an interview — not to trap, but to signal your awareness.
EU AI Act, Art. 5 — bans automated emotion inference in employment contexts (artificialintelligenceact.eu).
EU AI Act, Art. 50 — mandatory transparency: a generative AI system that interacts with you must be identified as such.
GDPR, Art. 22 — right to human intervention for any significant decision based solely on automated processing (CNIL).
30-day pivot plan (May-June 2026)
Thirty days is short. It's also enough if you calibrate well.
Days 1-10 — Audit + first prompts. Lay out what you've got: cases played, PEI written, frameworks mastered. Identify gaps. Run 5 prompting sessions on a deliberately throttled ChatGPT (system prompt: "answer with vague numbers, no sources"). Get comfortable with the discomfort.
Days 11-20 — 8 hybrid mock cases. You run with a human partner who poses the business question, you prompt the AI out loud, you review its output in front of them, you deliver your recommendation. Focus on judgment under fuzzy output — 25 minutes per case, on the clock.
Days 21-28 — PEI for the Lilli era. Rewrite your 4 core stories (Drive, Leadership, Personal Impact + a wildcard) around human-vs-tool arbitration scenarios. 3 to 4 minutes each. Record yourself.
Days 29-30 — Full McKinsey 2026 simulation. One complete cycle: Lilli-like prompt → critical review → structured recommendation → PEI. Debrief with a coach or via an AI platform.
On Velyq, you can practice this triptych on our AI interview platform with Lilli-style prompts built in. Also schedule an AI-era CV audit before your first round — the upstream filter has changed too in 2026.
FAQ
Is McKinsey actually using Lilli in interviews in 2026?
Yes. Delphine Zurkiya (Senior Partner, McKinsey) confirmed it to Business Insider in January 2026. Progressive rollout in final rounds, including European offices.
Is the classic case completely dead?
No, but it's become insufficient. Early rounds remain framework-friendly; late rounds embed Lilli plus judgment under ambiguity.
Can I use ChatGPT during a McKinsey Zoom case?
No. Marc Cosentino has documented cases where the interviewer detected it immediately and the candidate was blacklisted cross-firm (McKinsey, BCG, Bain).
What's the difference between Lilli (McKinsey) and Casey (BCG)?
Casey = standalone BCG chatbot, neutral screening. Lilli = embedded in a human-interviewer round, late-stage McKinsey.
Do I need to be an AI expert to ace the McKinsey 2026 interview?
No. The Guardian states it plainly: "collaboration and reasoning, not technical AI expertise".
Has the PEI changed?
Yes, contextualized for the Lilli era: they want stories about your human-vs-tool arbitration.
What rights do I have in Europe if I'm assessed by an AI?
GDPR Article 22 plus AI Act Articles 5 and 50: right to human review, transparency about the system, ban on emotion inference in employment.
How many cases should I practice in 30 days to pivot?
Velyq target: 8 hybrid mock cases plus 4 rewritten PEI plus 1 full simulation across 30 days.
Is McKinsey still hiring in 2026 despite the layoffs?
Yes, but the target profile is shifting to "AI-fluent generalist" rather than "framework athlete". Amodei (Anthropic) talks about up to 50% of entry-level white-collar jobs at risk (Business Insider).
Which Velyq resource should I use to train?
Our AI interview platform for Lilli-style mock interviews, and the CV analysis tool to calibrate your file for the AI era.
Key takeaways
- The classic McKinsey case isn't dead, it's demoted: necessary but no longer sufficient.
- Lilli is officially in the late rounds since January 2026 (Zurkiya, Sternfels).
- You're evaluated on your judgment in front of deliberately vague output — not on your AI mastery.
- Using unauthorized AI = cross-firm blacklist, instantly.
- The PEI remains but pivots into AI-era mode: human-vs-machine arbitration.
- In Europe: GDPR Article 22 + AI Act give you a right to human review.
- 30-day pivot plan: audit → hybrid mocks → rewritten PEI → final simulation.
