DEPLOYEDAI-era interview training
RankBootstrapping
XP 0 / 250
0
0%
Ready
Applied AI / Forward Deployed Engineer · Round-by-Round Playbook · Lesson 4 of 4

The discovery-call simulation

7 min

A newer round — heaviest at Anthropic, spreading fast — puts you in a role-play discovery call: the interviewer plays a VP at a fictional enterprise; you have 30–60 minutes. Candidate reports call it one of the strongest offer predictors in the loop, and the failure mode is universal: pitching instead of discovering.

Run it like a real discovery call

  1. Open with their world, not your product: 'walk me through how this works today' — map the workflow and where it hurts.
  2. Ask about past attempts: 'what have you tried — including AI — and what happened?' Failed-deployment stories are gold.
  3. Quantify the pain: how often, how expensive, who feels it?
  4. Find the immovables: compliance, data residency, systems that can't change, security review, procurement.
  5. Identify the people: daily users, exec sponsor, likely blockers (IT, security, legal).
  6. Define success together: 'if this works, what number moves?' — and how do they judge quality today?
  7. Close crisply: summarize what you heard, state what you'd validate first, propose a concrete next step.
  • Don't architect in the first half. If pushed — 'so what would you build?' — earn it: 'a couple more questions so I don't design the wrong thing.'
  • Aim to listen more than you talk; take visible notes; use their vocabulary, not yours.
  • It's OK to not know: 'I'd want to see the data before promising that' is a senior answer in this round.
Questions that score'What does your team do today when the model gets it wrong?' · 'How would you evaluate whether this is working — what's the number?' · 'What broke in your last AI pilot?' · 'Who has to say yes for this to reach production?' Each one signals you've shipped into a real organization before.

◆ Lock it in

  • Discover, don't pitch: today's workflow → past attempts → quantified pain → immovables → people → a success number → crisp close.
  • Hold off on architecture; listen more than you talk; admit what you'd need to verify.
  • Ask about failed pilots and current eval criteria — the highest-signal questions in the round.
Feynman drill — say it out loudYou're 5 minutes into a discovery call with a VP of Support at an insurer. Say your next four questions, in order, and why.
Step 1 rate your confidence · Step 2 pick your answer
Ten minutes into a discovery-call simulation, the 'VP' asks what you'd build. Best response?
Step 1 — how sure are you?