Interview
Harvey AI FDE interview guide
Harvey Forward Deployed loops test whether you can land legal AI workflows under confidentiality, accuracy, and firm operating constraints — trust and citation discipline matter as much as model fluency.
Legal vertical seats punish casual hallucination tolerance. Interviewers listen for whether you design review gates and measurable quality before promising leverage.
Interview loop matrix
| Stage | What they probe | Format | Pass signal |
|---|---|---|---|
| Recruiter screen | Domain humility, coding bar, travel, client exposure | 30–45 min call | You frame delivery ownership without claiming to be counsel |
| Technical screen | Systems, RAG, APIs, debugging | Exercise or discussion | Grounded design under incomplete corpora |
| Legal workflow case | Scoping, matter types, review gates, stakeholders | Ambiguous scenario | One practice-group thin slice with human review |
| Trust / confidentiality deep dive | Data boundaries, citations, auditability | Design + critique | Confidentiality and provenance as design inputs |
| Behavioral / HM | Partner communication, conflict, ambiguity | Story-driven | Precise risk language; no overclaiming advice |
Grading rubric
| Dimension | Strong | Weak |
|---|---|---|
| Domain humility | Scopes engineerable workflows; defers legal judgment | Talks like unauthorized counsel |
| Confidentiality | Matter isolation, access controls, audit trails by default | Shared corpora across sensitive boundaries |
| Grounding discipline | Citations, source checks, explicit uncertainty | Fluent answers without provenance |
| Review gates | Human review on high-stakes outputs before wider use | Autonomous drafting with no escalation path |
| Scope control | One matter type / one practice group / one KPI | Firm-wide AI assistant day one |
| Communication | Partner-ready risk language; clear promotion criteria | Hides failure modes to protect the demo |
Red flags
- No human review path for high-stakes legal outputs
- Ignoring matter confidentiality / data boundaries
- Equating fluent drafting with verified accuracy
- Cannot define a measurable pilot for one practice group
Practice scenarios
- Pilot for contract review assistants in one practice group — what is in vs out of scope for 30 days?
- An answer cites the wrong clause confidently — how do you diagnose retrieval vs generation and gate rollout?
- Partners want firm-wide enablement tomorrow — what evidence do you require first?
Pair with Enterprise RAG, agent evals, Glean FDE, and the Translation Matrix.
7-day prep plan
- Day 1–2 — Grounded RAG + citation discipline drill
- Day 3 — Confidentiality / matter isolation design
- Day 4 — Full practice-group pilot case
- Day 5 — Eval gates for high-stakes outputs
- Day 6 — Behavioral: partner communication under uncertainty
- Day 7 — Mock loop; score on the rubric
Comp context
Directional TC discussions often land around $190K–$380K. Details: Harvey AI FDE salary hub. Peers: Sierra, Glean.
Related hubs
Jump across salary, interview, and role-comparison pages for the same decision path.