Built around a sample profile: an Academy grad with a CS degree, a fighter pilot and Weapons School graduate, separating soon. Same person, two wildly different PM roles. Same prep philosophy — different texture. Retrieval beats re-reading every time.
Five phases. Each has a learn step and a test-yourself step. Do the test steps out loud. Active recall under mild stress is the only thing that translates to interview performance.
Also: Questions to ask · Post-interview debrief · Morning-of checklist · Cheat sheet
The A-10 is the closest USAF analog to the problem Anduril is solving: a slow, survivable, sensor-rich platform optimized for close air support in high-threat, comms-degraded environments. You have operated in exactly the scenario Lattice is built for. You know what it's like when the data link drops, when the ROE packet is incomplete, when the JTAC is talking fast and the clock is moving.
Frame it this way: "I've been the end user of the system Anduril is trying to build. I've also been the Weapons School instructor who evaluated whether a solution actually worked under stress. My job here is to make sure what ships to the operator is what actually helps — not what looks good in the demo."
| What you did | PM translation |
|---|---|
| Wrote and defended tactical employment publications (TTPs) | Technical writing, requirements documentation, spec authorship |
| Evaluated student tactics under simulated combat pressure | User research, usability testing in high-stakes conditions |
| Coordinated multi-domain, multi-platform strike packages | Cross-functional program management across engineering, ops, and leadership |
| Ran post-mission debriefs to drive learning and process improvement | Sprint retrospectives, A/B testing, product iteration loops |
| Identified and codified failures before they became patterns | Risk identification, incident post-mortem, failure mode analysis |
This is not a throwaway line. Elite athletes in an individual/team hybrid sport (track is both) demonstrate: autonomous performance under pressure, coachability without ego, and physical commitment as a proxy for mental commitment. In interviews, it signals you are not afraid of being measured and you know how to train, not just perform.
Use it when asked about resilience, setbacks, or how you develop: "I competed in track at the Academy — that experience of training alone for months toward a single measured moment taught me how to build self-directed practice loops. I run my professional development the same way."
Most veterans who apply for PM roles can't engage with engineers technically. You can. Lead with it:
At Anduril specifically: you own a product from concept to operator hands. That means writing the requirements, working with engineering on feasibility, representing user needs in every sprint, briefing leadership on tradeoffs, and being accountable for what ships. You are not a project manager (that's scheduling). You are the person who decides what gets built and defends why.
The defense-specific addition: you must also understand the procurement environment (ACAT levels, FAR/DFARS, OTA contracts), the classification environment (CUI, SAP, ITAR), and the operator environment. The last one is your advantage — most PMs here come from commercial tech and have never been the user.
Lattice is Anduril's platform play. Rather than selling individual hardware, Anduril wants Lattice to be the operating system for autonomous defense — the layer that ingests sensor data from drones, towers, radar, satellites, and human observers, fuses it into a common operational picture, and enables autonomous or human-authorized actions.
Think of it as: if AFDATC, GCCS-J, and a modern ML inference engine had a child designed for contested, comms-degraded environments. For an A-10 pilot, this is the system that should have been giving you better SA over the target area. For Anduril's PM: you need to understand the data flow (sensor → fuse → display → authorize → act) and where the gaps are in current operator experience.
A PRD answers four questions: Who is the user? What problem are they having? Why does it matter (quantified if possible)? What does success look like (acceptance criteria)?
Structure: problem statement → user stories ("As a [user], I need [capability] so that [outcome]") → functional requirements → non-functional requirements (latency, reliability, security) → out-of-scope → success metrics.
Your edge: you can write the user story from memory. "As a CAS controller, I need the system to distinguish between a moving vehicle and a static one within 3 seconds so that I don't call in a strike on a broken-down civilian truck." That sentence — written by an actual A-10 pilot — is worth ten pages of user research from someone who's never been there.
Other Transaction Authority — a contracting mechanism that allows the DoD to bypass FAR/DFARS (the normal federal acquisition rules) for research, prototyping, and production of innovative defense capabilities. Anduril has used OTAs to move faster than traditional prime contractors.
Why it matters for PM: OTA contracts are often milestone-based ("deliver working prototype by X date, demonstrate capability, then negotiate production"). As a PM you'll be scoping those milestones and making sure engineering can hit them. Know the phrase; be able to discuss it at a high level.
Anduril's public position: humans authorize lethal action; the system enables faster and more accurate decisions. DoD Directive 3000.09 governs this — "appropriate levels of human judgment over the use of force." Anduril's products are designed for human-on-the-loop (system acts, human can override) or human-in-the-loop (system recommends, human authorizes).
As an A-10 pilot you have direct experience with ROE, positive ID, and the consequences of getting it wrong. This is your most important qualitative edge over a commercial tech PM: you understand the moral weight of the product you're building.
These are the questions Anduril is most likely to ask. Work through each one out loud, alone, before the interview. Don't memorize a script — build a clear structure and let the words come naturally.
Structure: who you are → what you did that's relevant → why you want this specifically.
Draft: "I'm a USAFA grad — computer science — and I flew A-10s for [X] years. I went through Weapons School, which is the Air Force's advanced tactics and instructor pipeline — essentially a graduate program in how to design and evaluate complex airborne operations. I've spent the last [X] years [brief post-separation path]. I want to be at Anduril because I've been the end user of the problem you're solving. I know what it looks like when a system fails the operator at the worst possible moment, and I want to be on the side that makes sure it doesn't."
Note: don't explain Weapons School defensively. Name it, let it land, move on. Anyone doing their homework will know what it is.
In the first 20 seconds: Does this person know why they're here? In the next 40: Is the military background a credential or a crutch? The Weapons School line either lands as elite signal or falls flat if over-explained. Say it once, connect it to product work, move. If they want more, they'll ask — that's a good sign.
"The Weapons School taught me that the most leveraged thing a tactician can do is not fly better — it's write better TTPs, so that everyone who comes after you flies better. PM is the same idea applied to product: the highest-leverage position for someone with my background isn't being the operator who uses the tool, it's being the person who determines what tool gets built, so that thousands of operators are better off."
STAR-L format: Situation → Task → Action → Result → Learning.
Frame: Think of any moment in the A-10, Weapons School, or your unit where you saw a procedural gap, a training deficiency, or a capability hole — and did something about it. Did you rewrite a TTP? Build a training scenario? Identify a mission planning tool limitation and propose a workaround? That is PM work.
Example structure: "During [deployment/exercise], I noticed that [specific user pain — e.g., pilots were spending 40 minutes pre-mission doing X manually when the system should have automated it]. I [action: wrote up the problem, built a workaround, coordinated with [relevant party]]. The result was [measured outcome]. What I learned was [insight about user needs / product design / cross-functional work]."
"I use a combination of impact and confidence. For Anduril's context: I'd start with operator lethality impact — does this feature increase the probability of mission success in a contested environment, or does it reduce operator cognitive load at a critical moment? Then I'd layer in technical confidence — can we ship this in a deployable state, or does it open a dependency chain that delays everything else? In the A-10 world we called this threat prioritization: not all threats are equal, not all threats are immediate. The discipline is the same."
Know the common frameworks: RICE (Reach, Impact, Confidence, Effort), MoSCoW (Must/Should/Could/Won't), and the simpler "what breaks if we don't do this?" test for defense contexts where the user consequence of failure is not a bad NPS score.
"I've spent my career in environments where the people I needed to execute for me could kill me with information I didn't have. You learn fast to listen before you decide. When an engineer pushes back, my first question is 'what do you know that I don't?' — sometimes they're right and the requirement needs to change. Sometimes the requirement is correct but the framing was wrong and I need to explain the operator context better. The debrief culture in aviation helps: we assume everyone in the room saw something useful, and the goal is to synthesize it, not to be right."
Pick one and be specific. Don't say "I'm open to whatever." Suggested answer for an A-10 background:
"Lattice's CAS integration layer — specifically the capability that connects JTACs on the ground to autonomous sensors in the air. That interface has been broken for decades: the data is there, the connectivity has gotten better, but the operator experience hasn't kept up. I've been on both sides of that radio call — I know what the JTAC needs and I know what the pilot needs. That's the problem I'd want to close."
Alternative: Roadrunner / counter-UAS if you have experience with the drone threat. Ghost-X if you want to focus on ISR product management.
Be specific. Be honest. Show learning. The debrief culture gives you a natural structure — pilots talk about failures every day, that's what the debrief is for. Use that frame:
"In [context], I made a decision to [action]. The result was [specific negative outcome]. In the debrief I identified that I had [root cause: anchoring bias / incomplete information / mis-prioritized / failed to delegate]. What I changed was [specific behavioral change]. The next time [similar situation] came up, I [applied the learning]."
Don't pick a story where the failure was small or where you're actually secretly the hero. Pick a real one. Anduril is a culture that values honesty over polish.
"I use the cost-capability-schedule triangle and I anchor it in the operator consequence. 'If we add this capability, we delay by three weeks and the system ships after the exercise window. If we cut it, the operator loses [specific thing]. Here's what I recommend and why.' The non-technical stakeholder doesn't need to understand the code — they need to understand the tradeoff in terms they can defend to their chain. I brief it that way."
Don't be vague. "I want to own a major product line — something that's in the hands of real operators in real operations. I'm interested in the intersection of autonomy and human judgment, and I think that problem is going to get harder before it gets easier as the threat environment evolves. In five years I want to be the person who's shaped how Anduril has approached that — with a track record of products that actually work in the field."
Rate yourself 0–2 on each dimension. Target: 8+ total. One rep won't tell you much — run three reps of the same question and watch the trend.
| Dimension | 0 — Weak | 1 — Solid | 2 — Strong |
|---|---|---|---|
| Substance | Vague, no specifics | One concrete example | Specific, quantified, unexpected detail |
| Structure | Rambling, no arc | STAR shape but loose | Clean STAR-L, under 90 sec, no filler |
| Relevance | Doesn't address what they're probing | Connects, but loosely | Directly hits the competency being tested |
| Credibility | Sounds rehearsed or generic | Sounds real, feels earned | Detail only someone with your exact background could give |
| Differentiation | Could be any veteran candidate | Has a military angle | The A-10 / Weapons School angle is the story, not an add-on |
| Dimension | Anduril | Lockheed / Raytheon / L3 |
|---|---|---|
| Development speed | Weeks to prototype, OTA contracts, spiral development | Years to milestone, FAR/DFARS overhead, ACAT II/I programs |
| Culture | Silicon Valley talent + veteran operator mentality; flatly structured | Matrixed, process-driven, risk-averse, institutionally slow |
| Operator relationship | Direct: PMs and engineers embedded with units | Mediated: program office → contractor → LRIP → fielding |
| PM role | Own the product end-to-end, write requirements, brief leadership | Program manager (schedule/budget focus) + systems engineer (requirements) |
| Risk tolerance | Ship early, learn from operators, iterate | Qualification before fielding, low tolerance for early failure |
1. Anduril is building a new capability to help A-10 pilots (before retirement) and their eventual replacements coordinate CAS with ground forces in a GPS-degraded environment. You are the PM. What's your first move?
Think: user research → problem definition → MVP scope → success metric. Reference your own experience as the user.
2. Engineering tells you that the feature the JTACs most want (real-time blue force tracking on Lattice, visible on the pilot's display) will delay the release by 6 weeks. The PM director wants to ship on time. What do you do?
Think about: what information you'd gather, what the operator consequence of each option is, how you'd frame the recommendation, and what you'd do if overruled.
3. You've been given a new product area: Roadrunner's human-machine interface for the launch operator. You have no existing requirements, one engineer, and a demo in 90 days. Walk me through how you approach this.
Think: what you'd do in week 1 (user interviews, existing pain points), what an MVP HMI looks like, what success at 90 days means vs. what a production-ready system looks like.
Don't jump to solutions. Ask: "What specifically are you doing when it breaks down?" → "How often does this happen?" → "What do you do instead right now?" → "What would a good day look like?" You are mining for the specific behavior, not the vague sentiment. "I hate the interface" is data; it's not requirements. Your job is to convert it.
First: believe them and get specific. "What exactly makes it impossible — is it compute, latency, data availability, or time?" Then: understand if it's impossible given current architecture or impossible ever. Then: go back to the operator need. Can a different implementation satisfy the same underlying need? Write the requirement to the capability ("the system shall enable the operator to positively identify a moving target within 5 seconds"), not the implementation ("the system shall use ML fusion layer X"). If it's genuinely impossible, escalate honestly with a proposed alternative.
Three most common failure modes: (1) training gap — they don't know how to use it, (2) fit gap — it solves a problem they don't actually have, (3) trust gap — they tried it once, it failed, now they default to old behavior. Go visit the unit. Watch them work. Don't send a survey. Observe the specific moment they switch away from your product to the old tool — that moment is your requirements document.
Have a real answer. Something specific, grounded in your operator experience. Example: "The biggest gap I saw in the A-10 is the interface between the kill chain decision and the comms-degraded environment. Everyone assumes the pilot has situational awareness — but the moment the link drops, you're flying on last-known positions. I'd build a Lattice capability that predicts threat movement in the 30-second comms window — not to replace judgment, but to sharpen the estimate the pilot already has to make." Confident, specific, operator-grounded, connected to Lattice.
Good questions signal you've done the work. These are calibrated for an Anduril PM interview.
Treat it like a mission debrief. Write your answers while they're fresh.
1. What question did they ask that you weren't ready for? What's the better answer?
2. What answer landed well? What made it work?
3. What did you learn about the role or the team that you didn't know before?
4. What's your honest read on fit — theirs for you, yours for them?
5. What do you do in the next 24 hours?
[Interviewer name] —
Thank you for the time today. [One specific thing from the conversation that was interesting or that you're still thinking about.] It reinforced why I'm excited about this role specifically.
[One sentence that connects your background to something they said — not generic, not a summary of your resume. Prove you were listening.]
Looking forward to next steps.
[Your name]
Rate each story: ★★★ Strong Fit · ★★ Workable · ★ Stretch · — Gap. Pre-fill so you're not constructing under pressure.
| Your story | Tell me about yourself | Why PM? | User need → solution | Prioritize features | Failure / learning | 5 years? |
|---|---|---|---|---|---|---|
| Leadership: [WS instruction / flight lead] | ||||||
| Technical: [CS project / JTAC problem solved] | ||||||
| Failure: [specific, honest, clear learning] | ||||||
| Innovation: [TTP rewrite / tool you built] | ||||||
| Cross-functional: [multi-domain coordination] |
Same operator. Wildly different shape. Where Anduril hires you because you were the customer, Amazon hires you because you can become one.
Amazon has 16 LPs. Three of them are where vets win and three are where vets get rejected. Pre-write a STAR-L story for each.
Each story should be 90 seconds, tagged to 1–2 LPs, and end with a one-line Learning. Write them out before you walk into a room.
Amazon often gives PM candidates a Working Backwards prompt. You have 30–60 minutes to draft a 1-page PRFAQ (Press Release + FAQ) for a hypothetical product. The press release goes first — written as if the product already launched.
Notice the trap: you'd want to build the most complete product. Don't. Pick one user, one need, one MVP. Customer obsession is "deeply understand this user," not "serve everyone." Cut your scope. Write the press release for that minimum. The FAQs are where you show range.
| Dimension | Anduril | Amazon |
|---|---|---|
| Why they want you | You were the customer. You know what works under fire. | You ship results under pressure. You can become a customer expert fast. |
| The trap | Becoming the "war story guy" who can't write a spec. | Leading with rank or jet. Amazon doesn't grade those. |
| Format | Behavioral + product critique + a Lattice / autonomy hypothetical. | STAR-LP + Working Backwards + writing sample + Bar Raiser. |
| Vocabulary | JTAC, ROE, denied environment, kill chain. | Customer, mechanism, single-threaded leader, two-pizza team, working backwards. |
| What “ownership” means | You own the engagement. | You own the metric, the doc, the launch, the rollback, the post-mortem. |
| Compensation tier | Mid-six figures, light equity. | Top-tier total comp, RSUs vesting heavily in years 2–4. |
The Anduril and Amazon workbooks above were built using the protocol on this page. You can build one for any role you're interviewing for in about 30 minutes. Here's how.
Before you open Claude, collect these into a single document:
Most candidates read the JD once and move on. Work through it with these six lenses — mark what you find. Give your notes to Claude alongside the raw JD.
Words or phrases that appear more than once. Repetition is emphasis. If "cross-functional alignment" appears three times, that's not an accident — they have a problem there.
What's in the first bullet? The last paragraph? First item in a list gets the most weight; last item is memorable. Middle items are filler.
Separate the "must have" from "preferred" language explicitly. Candidates over-index on preferred items they don't have and under-index on non-obvious required items they do.
"Own" vs. "support." "Define" vs. "contribute." "Lead" vs. "partner." Verbs tell you the actual power and scope of the role more than the title does.
What problem is this role solving for the company? If they're hiring a PM for "operator-embedded product development," they probably have a gap between engineering and the field. Name that gap to Claude.
What a JD doesn't say is as informative as what it does. No mention of roadmap ownership? They may have a design partner or above-PM who controls the roadmap. No comp transparency? Ask in the screen.
Paste this into Claude after pasting your inputs. Edit the role/company name to match yours.
Claude's first draft will be 80% there. The remaining 20% is the conversation. Three follow-ups worth running:
Once you've got a workbook you like, save the conversation. The next role you interview for, paste in the new JD and ask Claude to produce a new workbook in the same format. The protocol amortizes — your second workbook takes 10 minutes, not 30.