Day-Of Interview Tactics
Read the morning of each round. Pre-loop checklist, structural moves, traps to avoid, recovery patterns, and the questions to ask them at the end.
The night before
- Reread 00-START-HERE, 01-the-role, 02-positioning, and the questions in 15-interview-questions.
- Skim 14-domain-context — Workday and ATS terms cold.
- Sleep. Caffeine recovers from one thing; sleep does not.
- Pick a quiet room with good camera, light, and audio. Test the call link.
- Pull up the JD in a tab; print a copy if you want it physical.
- Have water, notepad, pen. Phone off the desk.
30 minutes before
- Re-read your one-paragraph self-summary out loud (from 02).
- Re-read your three pre-loaded stories. Once each. No new content now.
- Stand up, walk for 10 minutes. Don't keep cramming.
- Three minutes before, sit, slow breaths for 60 seconds. Smile — it changes your voice.
Pre-loop checklist (do once before round 1)
- Know who interviews you. LinkedIn-skim each person. Note their background — Workday-heavy vs. AI-heavy. Calibrate which depth to lean into.
- Decide your archetype. Are you AI-strong / HR-light or the reverse? See 02. Don't switch mid-loop.
- Pre-load three concrete stories. Said-no story. Multiplier-handoff story. Incident story. 90 seconds each.
- Pre-load three questions per round. See below.
- One-paragraph self-summary memorized to fluency, not by rote.
- Closing line — see below.
During — structural moves
Open every answer with a 5-second framing
Before diving in, frame:
- "Two things matter here..." (signals structure)
- "I think about this in three buckets..." (signals structure)
- "Short answer is X. The longer version..." (value first)
- "Depends on the risk tier — let me cover T1 and T3 separately..." (judgment)
- "Let me restate to confirm — you want me to..." (clarification)
This buys you thinking time and signals you're not blurting.
Bring the failure mode unprompted
For any design answer, end with "the failure mode I'd watch is X, and the mitigation is Y." Senior signal.
Reach for the right nouns
Effective-dating. Position vs Job vs Job Profile. EIB partial failure. Sup org. Calc field. Payroll lock window. Process Owner. Risk tier. Idempotency key. HITL gate. The vocabulary signals fluency.
When you don't know something
Workday and AI engineers detect bluffing instantly. The cost of a caught bluff is much higher than the cost of a clean gap.
Patterns that work:
"I haven't worked directly with X. My closest experience is Y. Want me to reason about X from first principles, or pivot to the related thing I have done?"
"I understand the shape conceptually but haven't run one in anger. Tell me what specifically you'd want me to dig into."
"I'd reach for the reference for the exact syntax / function name, but the pattern is..."
When stuck on a design problem
Three moves:
- Risk-tier it. "Is this T1 or T3? That changes the answer." Often, just naming the tier creates structure.
- Decompose into read / propose / commit / audit. The four layers are always present. Walk through each.
- Name the constraints first. "Three constraints shape this — payroll calendar, PII residency, audit replay. Given those..."
If still stuck, ask: "Let me make sure I'm answering the right question — are you most interested in the failure modes, the data flows, or the rollout?" They'll often refocus you usefully.
When you need a moment
- "Let me think about that for a second." Five seconds of silence is fine. Don't fill with words.
- Sip water. Look at notepad. Reset.
- "To make sure I cover the right thing — three buckets to think about: [list]. Want me to go deepest on one?" Buys time + signals structure.
When they push back on your answer
Two flavors of pushback. Both are good signals — they're engaging.
- "What if X instead?" — they're stress-testing. Engage on the merits. "That'd shift the trade-off in [direction]. The case for it is [Y]. The case against [Z]. In this domain I'd still favor [original], because…" If you genuinely change your mind, say so. "You're right — I hadn't considered [Y]. That changes my answer to [Z]." Updating in real-time is a strong signal.
- "Hmm, I'd worry about..." — they see a hole. Listen carefully. Name the hole back. "I hear the concern about [thing]. The way I'd mitigate it is [X]. If that doesn't fully address it, the next layer is [Y]."
Vocabulary moves — the small things that signal seniority
- Say "propose, not commit" when describing agent writes.
- Say "effective-dated" when describing Workday state.
- Say "data classification" when describing PII handling.
- Say "named Process Owner" when describing ownership.
- Say "halt criteria", not "errors."
- Say "audit log is the regulator's view".
- Say "idempotency key on commit" when describing retries.
- Say "sampled review" for low-tier flows.
- Say "two-key approval" for severance / equity / cross-functional.
- Say "approve-with-diff" for high-tier HITL UIs.
Specific traps to avoid
You'll be asked something like "can the agent just do the comp change?" The answer is never just yes. Always: "For payroll-adjacent actions, the agent proposes, a human approves with diff. The cost of full automation here is paid in payroll exceptions and trust." If you say "yes, the agent commits the Workday write," you've failed the round.
When the conversation hits cross-jurisdictional PII, don't try to sound lawyerly. Don't quote articles. Do say: "I'd flag this for Legal and Privacy review. I won't unilaterally interpret transfer mechanisms. Operationally I'd default to data-minimization and pseudonymization wherever I can."
If you haven't run an EIB, say so. If you don't remember exact calc-field syntax, say so. The vocabulary is dense enough that fakers are spotted in 90 seconds.
Don't decompose every workflow into 5 agents. Default to single-agent + many tools; decompose only when you can articulate why (different models, different evals, prompt-budget exceeded). Over-decomposition reads as "wants to look impressive."
"Reduce manual work by 80%" without context is a flag. Better: "My north-star outcome is edit rate trending down and Process Owner sign-off that they've extended a workflow themselves. Hours saved is downstream of those."
"I'd build it and ship it" misses the multiplier point. Always: "I'd build with the Process Owner from day one — they're the long-term owner, not me."
Questions to ask them — at the end of each round
For an AI / engineering interviewer
- "Where do n8n, custom Python, and MCP servers cleanly divide responsibilities in your existing builds — and where is that boundary still being figured out?"
- "What's your eval discipline today? Do you have labeled sets that meaningfully gate promotion to production?"
- "When an automated change to a worker record is later questioned, what does your replay/audit story look like?"
For an HR / People Ops interviewer (HRBP or HRKX)
- "Which workflows do you wish were automated but haven't been? What's blocked them — risk, capacity, ownership?"
- "What does the first month of working with the AI architect look like from your seat? Where do you expect to be self-sufficient by month six?"
- "What does Payroll Ops need to feel comfortable with an agent in their world?"
For a Workday / integrations interviewer
- "Which Workday integration patterns are in production — Studio, Core Connectors, RaaS, EIB — and which gives you the most pain?"
- "How do you currently handle changes during sandbox refreshes? What breaks regularly?"
- "Are you using Prism, and if so, what does your warehouse split look like — Prism, BigQuery, both?"
For a leadership / hiring-manager interviewer
- "What's the highest-stakes People workflow currently being considered for AI assist? Where is the human approval gate non-negotiable?"
- "What does success in 6 months look like? In 18 months? Who's going to be telling me at month 12 that this is working?"
- "Where would a new architect have the most leverage in their first 90 days?"
Your closing line
Have one ready. When asked "anything to add?" or "any closing thoughts?" — deliver this in under 30 seconds, conversationally.
"The part of this role I find most interesting is that the constraints — payroll-adjacent HITL, employee PII, cross-jurisdictional privacy, the halt-or-redesign bar — force good architecture. I'd rather build AI in that environment than somewhere autonomous demos are the bar. If you decide to bring me back, that's the work I'd want to do."
If something goes badly
You blanked on a question. You said something wrong. You ran over time. Recovery patterns:
- Acknowledge briefly, move on. "That answer didn't land for me. Let me try the next question and come back if there's time."
- Re-attempt with a frame. "Let me retry that with a cleaner structure — three things matter..."
- Don't dwell. One bad answer doesn't tank the loop. Three bad answers + visible spiraling does. Stay calm.
Between rounds: stand up, walk, drink water, reset. Don't re-litigate the round in your head.
After the interview
- Within 24 hours: send a short thank-you to your recruiter or hiring manager. One paragraph; reference one specific thing from the loop ("I enjoyed the discussion on payroll-calendar-aware change control"); confirm interest; close.
- Don't fish for feedback in the thank-you. If they offer, great. If not, ask later if you want to.
- Write down what you'd answer differently next time, regardless of outcome. Loops are reps.
The one mantra
Honest. Prepared. Fast learner.
That's the posture. The content is already in this folder.
You're closer than you think. Go.