Day-Of Tactics

Pre-loop checklist

Night before

• Reread chapter 01, 02, 14 (vocab cram sheet).
• Reread the JD once. Note the five responsibility areas, the two horizons.
• Sleep. Don't cram new material at 11pm.

90 minutes before

• Re-skim the vocab cram sheet aloud (14 one-pager).
• Re-skim chapter 04 — whiteboard template.
• Re-skim chapter 11 — coding template.
• Look at your one-line opener (below) until it's reflexive.

15 minutes before

• Glass of water. Camera framing. Mic check.
• Note pad open with: the five JD responsibility areas, the risk tiers, the systems list, your opener.
• Read the "right after" section now so you remember to do it later.

The first 60 seconds

The "tell me about yourself" answer. Memorize this shape from chapter 02:

That sets the frame, names your strength, names your gap, signals honesty, hands back control. The rest of the loop runs more comfortably from there.

The structural moves — every round

Senior candidates do these on every type of round.

1. Clarify before answering. "Before I dive in, let me confirm the question I'm answering." One sentence; then proceed.
2. Decompose. Whatever the question, identify 3-5 sub-parts. "There are three pieces here: A, B, C."
3. Design. Walk the chosen direction explicitly. Use the JD vocabulary (risk tier, Process Owner, MCP, HITL).
4. Tradeoffs. Name what you didn't pick and why.
5. Failure modes. "Three things that go wrong: hallucination, idempotency violation, integration drift. Here's how the design handles each."
6. Test & rollout. "Evals are the change gate; rollout is sequenced; first version is Tier 1."

That sequence — clarify, decompose, design, tradeoff, failure modes, test — works for design rounds, coding rounds, and behavioral rounds. Make it muscle memory.

The design-round shape

Reproduced from 04 because you'll want it here too. 20 minutes:

1. Clarify scope (2 min) — entities, cadence, systems, must-have output.
2. Risk tier & Process Owner (2 min).
3. Decompose into 4-8 steps (3 min); LLM-driven vs. deterministic.
4. Pick a pattern (1 min) — single-agent, planner/worker, router; defend it.
5. Tools & MCP (4 min) — list them; read/write split.
6. HITL gates (2 min).
7. Audit & rollback (2 min).
8. Evals & rollout sequence (2 min).
9. Three failure modes (2 min).

The coding-round shape

1. Restate the problem in your words.
2. Clarify: size, sorted, duplicates, currency, decimal precision, empty input.
3. Walk one example on the shared doc.
4. Brute force + its complexity.
5. Name the pattern — hash, two-pointer, prefix sum, sliding window, topo, sampling.
6. Improved complexity claim before coding.
7. Code with narration. Keep talking.
8. Edge cases: empty, single, duplicate, very large, currency mismatch, Decimal vs float.
9. Test trace: walk a small input through your code.

The behavioral-round shape

Standard STAR — Situation, Task, Action, Result — but with two tweaks for this role:

• Lead with the constraint. "The thing that made this hard was [SOX / regulated / data sensitivity / scale]." That signals you understand the world this role lives in.
• Finish with the artifact. Not just "we shipped it" — "we shipped it, and the audit log / runbook / eval suite that came out of it is still the platform default."

Have these stories ready:

• A time you halted a build.
• A time an agent hallucinated and you caught it.
• A time you said no to a stakeholder without losing them.
• A time you ramped quickly on a new domain.
• A time you worked across functions (eng + finance / eng + legal).

Traps to avoid

Graceful recovery

You will say something imperfect. That's fine. The recovery shape:

1. Notice. ("Actually — let me revisit that.")
2. Name what you'd change. ("Re-reading what I said: that wouldn't work because X.")
3. Replace it cleanly. ("Better answer: Y, because Z.")
4. Move on. Don't dwell.

Senior interviewers like watching this — it shows you're calibrated and not defensive. A candidate who corrects themselves smoothly reads stronger than a candidate who got it right by chance.

Other recoveries:

• Blanked on a term: "I'm blanking on the exact word, but the concept is — [describe it]. The word is on the tip of my tongue." Often the interviewer fills it in; even if not, you've shown the substance.
• Wrong direction in coding: "I've spent 5 minutes on this; I don't think this approach scales. Let me restart with [pattern]."
• Don't know an answer: "I don't know. Want me to reason about it from first principles, or would you rather move on?"

Questions to ask them — by round

Have 2-3 ready per round type. Pick from this menu:

To the hiring manager

1. "What's the highest-stakes finance workflow currently being considered for agentic automation? What gates are non-negotiable?"
2. "How does the Automate Everything CoE actually govern? Who has veto power on a Finance build?"
3. "What's the biggest unsolved problem in close right now? Where does a new architect have the most leverage in their first 90 days?"

To an engineering peer

1. "Where do n8n vs. Python vs. MCP cleanly divide responsibilities in your existing builds, and where is the boundary still fuzzy?"
2. "How is your eval harness today? What would you change about it if you could?"
3. "What does the audit-log substrate look like? Indexed by what?"

To a finance peer (Controller / Treasurer / FP&A)

1. "Walk me through the most painful manual process in your week. Where would an agent help, and where would it actually make things worse?"
2. "When external auditors walk through an automated control, what does the replay story look like today?"
3. "What's a build you've seen that failed in this space, and why?"

To someone senior (VP / Director)

1. "On the two horizons — deploy-now and redesign-the-architecture — which has the most pressure on it in the next 12 months?"
2. "What's the bar for 'platform' rather than 'scripts' in your view?"
3. "How is success in this role measured at the 6-month mark, and at the 18-month mark?"

Closing statement

End every round and the loop overall with a clean close. Template:

The "offer the artifact" line is high-signal. It signals that you already have artifacts, and it gives the interviewer a graceful next step.

Right after the loop

• Write down, within 10 minutes, what each round asked and how you answered. Memory decays fast.
• Note which round felt weakest and which felt strongest. Both inform the follow-up.
• Send a thank-you to the recruiter and (where appropriate) the hiring manager within 24 hours — brief, specific, and offering one concrete artifact if they'd like it.
• Whatever the outcome, you got better. The next loop starts from a stronger floor.