Operating Under Fire

I’m prepping for a panel where the first case is a constructed crisis. A doc lands describing seven simultaneous fronts in week 4 of a Shape Up cycle, six weeks into a hypothetical Director of Engineering tenure, and asks for a 30-day plan, what I won’t do, a bad-but-necessary call, and what I’m missing. The trap is to engage at the level of the prompt. Answering “give me a 30-day plan” with a 30-day plan is the candidate move. The director-level response rearranges the prompt before answering it.

What the prep keeps returning to is that the named fronts are symptoms, not the work. The substrate underneath — culture, delivery standards, cross-functional norms, observability — is what determines whether the same list shows up again in 90 days. The panel is testing whether I see that layer, not whether I can solve seven problems on time. What follows is the framework I built. It generalises: any time you join a fire mid-blaze, lose a director and step in, or run a post-incident rebuild, the same shape applies.

Rank by failure cost

Urgency drives reflex; failure cost drives priority. They diverge often. The thing firing tomorrow can be smaller than the thing firing in eight weeks.

In my case, the demos were urgent (next week) but recoverable. The audit gap was weeks away but existential — losing Class IIb SaMD certification ends the EU business. Reordering the seven fronts by worst-case-if-I-fail surfaced two items at the top that the panel had listed in the middle, and dropped two items down that were day-of fires.

The exercise is short. Worst, then less-worst, all the way down. If you can’t articulate why one is worse than another in one sentence, you don’t actually believe the order yet.

I do, I ask

For every front, separate what only you can do from what you’d ask of named owners. A table works. Three columns: front, “I do”, “I ask”.

The “I do” column is the test. If it’s the entire row, you haven’t delegated. If it’s empty, you’re abdicating.

Most fronts in my map had small “I do” columns and large “I ask” columns. A few were inverted — scope hammering with cross-org outreach to repair Sales/CS trust, the demo reset back to the CEO/CTO, the prioritization call on audit capacity. Director-only because the signal of who’s deciding is part of the work. An EM saying “demos won’t include feature X” gets escalated around. A director saying it lands.

A fourth column helps: who to signal. Same person isn’t always involved at the same depth. Naming the broadcast list per front prevents the “wait, why didn’t you tell me” conversation later.

Operating in async, with scheduled anchors

The naive calendar reads as N fronts × M touchpoints, which inflates a week to 30+ scheduled hours. That math doesn’t survive contact with a 55-person company anyway, because the actual rhythm is async. Slack threads, hallway-equivalent DMs, the half-thought your CTO drops at 11pm.

What scheduling buys is spine. Two anchors a week with the load-bearing peers — for me, the QA director and the CTO — give the async rhythm landmarks. Threads roam during the week and converge at those points. The discipline isn’t to consolidate fifteen pings into five conversations. It’s to layer a few real conversations on top of the pings so the pings have somewhere to land.

The CTO note matters in particular. At a 55-person Series A with a co-founder CTO, the weekly 1:1 is a punctuation mark in a thread, not the relationship. Stop treating the CTO as a manager you sync with. Treat them as a peer who’s in the work. Your job is to be a force multiplier on engineering operations so the CTO can spend their attention on the strategic and customer fronts where they’re the irreplaceable one. Practically: be reachable on Slack across the day, surface decisions early so the CTO doesn’t surprise you with a customer commitment that breaks scope, volunteer information without waiting to be asked. The synchronous 1:1 is for the harder stuff.

The honest week 1

Week 1 isn’t thirty hours of scheduled conversations. It’s fifty hours of which thirty are scheduled, and the rest is Slack, codebase reading, fires that weren’t on the doc, and the side conversations where things actually move.

Realistically: 30–40% of the calendar slips, at least one EM isn’t ready to absorb delegation, a stakeholder escalates over your head once, the cycle closes messy with one bet visibly missing, on-call goes live a week later than the deadline, the audit evidence package slips by a week, an unnamed eighth thing fires. It always does. Day 30 is the diagnostic moment, not the delivery moment.

Inside that fifty hours, about five protected slots have to land for the plan to mean anything. Some on the calendar, some on a Saturday morning, some on a flight where the WiFi finally fails. Three buckets:

• Written artifacts I author: re-set messages, policy docs, MR norms, betting-table preconditions. Writing in your voice and signed by you, even if collaborators draft.
• Decisions I make alone: prioritization calls, architectural sponsorship, EM-readiness map. Quiet thinking, then communicated.
• Sense-making: a director-level theory of the system. Which fragmentation gaps to attack, in what order. This one stubbornly refuses to land on a scheduled block. It shows up in journals, on walks, in the shower.

The theory-of-the-system slot is the most under-priced. Easiest to skip, biggest compounding return. Skip it and at day 30 you look competent and busy but you don’t have a thesis. The CTO will notice the absence by day 60.

The substrate the panel didn’t ask about

The panel’s named list is the symptom layer. There’s a parallel list — what you’d diagnose yourself — that the panel didn’t ask for.

In my case: a department health pulse, an interrogation of the 0-bug policy that came up in a previous interview (every named front bends if that policy bends), the AI-generated-code provenance question for SaMD paths, the cross-functional connective tissue that lets enterprise tenders not blow up the cycle, an observability layer for myself so I don’t have to dive into MRs to know what’s happening.

These don’t replace the named fronts. They run in parallel. The named fronts are the visible 30-day work; the self-found list is what determines whether the same panel writes the same case in 90 days. Surfacing this during the walkthrough signals you see the substrate, and it’s the connective tissue between Case 1 (operational) and Case 2 (year-1 reflective).

Believe the stated culture priority

If the hiring conversation surfaced a stated #1 priority — for me, the CTO saying culture was top of mind — believe it. It’s not an afterthought theme on a 12-month list. It’s the substrate every operational theme expresses.

The instruments of culture are the load-bearing items. Product engineering culture, performance management, delivery standards, decision-making norms, knowledge spread, leadership development. Every named operational theme — production discipline, regulatory-as-fabric, platform for scale, cross-functional contract — is downstream of one of those instruments.

Every named front becomes an opportunity to install a cultural norm. The manner of handling each front is the culture work.

Year 2 needs explicit moves

A director who arrives with operational competence but no thesis becomes a senior EM with budget. The thesis lives at Year 2.

What it looks like, concretely, at this scale is three or four explicit moves with capital, talent, or org implications. From my prep:

• Org design: do the four TL areas (Backend/Integrations, Frontend, AI, Engineering Excellence) survive a 14 → 25 engineer transition, or does the AI group spin out as its own squad with a dedicated EM by Q3?
• Build vs buy: the eval platform for LOLA. Build it ourselves, integrate a vendor like LangSmith or Braintrust, or hybrid? Each option has audit implications under Class IIb.
• Talent: when does formal levelling get installed — at 20 engineers, 25, never? My current position is to defer until 25, with promotion decisions CTO-aligned case-by-case until then.
• Methodology: does pure Shape Up survive the enterprise-integration mix, or does a parallel solutions-delivery track emerge with its own cadence by Year 2?

Three or four positions, not a long list. They’re claims you can be wrong about. They get tested, refined, sometimes abandoned. The point is to have them in writing where they can be argued with.

Make the panel a working session

Bring draft 30-day success criteria into the panel, hand them out, ask the panel to amend. This converts a one-way audition into a co-created contract. If they engage, you walk out with shared criteria. If they don’t, you’ve named the criteria that will define the first 30 days regardless.

Same logic on per-panel-member prep. Map who’s likely to ask what, and also what you want each person to ask. The first is preparation; the second is strategy.

A small adjustment that costs nothing: same person, different vocab depending on role. The EM reads engineering-specificity as the signal of whether you can run the function. The Chief of Staff reads cross-functional trade-off framing. The QA peer reads regulatory-quality fluency.

Pre-mortem the year-2 case

The 12-month reflective companion is cleanest as a pre-mortem. Imagine you failed at 12 months. What happened?

Sharper teeth than “what’s still not working.” What-went-well emerges by contrast. Year-2-defining items become “what I’d resolve to do differently.” The brief usually says they want unfinished, not a victory lap. Take that seriously — most candidates don’t.