Close or upgrade (months 6–9)

When I enter: chapter 3 exit gate is honestly green (Stabilize). I have charter wins, a signal artifact, and warm threads — not hope. Those threads have been alive since the war room.

Success (any one counts):

• Offer from Anthropic / OpenAI enterprise (or Applied / FDE-equivalent at those labs)
• Offer from FAANG AI Platform / Agent / Applied AI Staff+
• Offer from peer large-enterprise Principal / Sr Staff AI Platform
• Or leveled scope/comp on the bridge seat with a dated next gate — not infinite “maybe later”

Mission

Close the north star — or upgrade the bridge with leverage. Same pitch. Deeper war stories from the seat. Ruthless postmortems on rejects. Parallel loops when I can. Serial loops are how offers expire alone.

Chapter 3 signal
  → referral-first loops
  → screens → HM → onsite (isolation, evals, leveling, values/judgment)
  → compete / negotiate with real written offers only
  → sign north-star OR bridge upgrade with dates

Month-by-month

Quota

Employed:

Full-time search: 6–10 high-signal actions/day like the war room, but only north-star + peer Principal targets.

Track on the scoreboard.

What these loops actually probe

Expect depth — not slogans — on:

1. Multi-tenant isolation — threat model → layers → negative evals → Demo vs Strict
2. Eval false confidence — CI gates, suite kinds, what I still don’t measure
3. Gateway / HITL on irreversible tools
4. Access-before-ranking RAG — decline paths, not only ranking tricks
5. Org reach — standards, mentoring, ADRs, adoption by teams that don’t report to me
6. Stack honesty — Java/Spring enterprise history + Python agents; don’t overclaim daily Go
7. Judgment / values (especially lab Applied) — where I refuse automation; how I update under uncertainty

Verbal heroes remain AegisAI + Enterprise RAG. Supporting: VAP, Content Factory, Lucid outcomes, charter wins.

Applied / forward-deployed screens care more about discovery under a skeptical stakeholder, eval design from a vague ask, and concrete deployment risk than about a beautiful multi-region diagram. If I sound like a generic “AI platform architect” who has never sat with a scared enterprise buyer, they’ll feel it.

Scripts

“Why us / why now?”

I’m not hunting logos. I want a seat where agents and retrieval are the product — tools that can do damage, evals that can fail a build, honest tenancy. I did the homework on [specific public signal]. I’m here to see if the ownership and the leveling match that story.

I do the homework. “Because you’re a leader in AI” is noise.

“Design multi-tenant agents for us” (first 60 seconds)

I’d start with the threat model — cross-tenant read, tool invocation as tenant A against tenant B’s data, prompt injection that tries to confuse identity. Default is fail-closed: authZ before retrieval, tenant on every tool call, negative tests in CI, and a Strict mode that doesn’t pretend Demo SSO is production. Happy to whiteboard layers next — and I’ll mark what my public Strict path still doesn’t cover.

“How do you think about safety?” (no sermon)

I treat irreversible actions as the hard boundary — publish, refund, CRM write, deploy. Those go through policy and often a human. Soft failures — wrong answer with a citation — get evals and monitoring. I’m suspicious of fluency. If we can’t name the deny path and the escalation path, we don’t have safety. We have a demo.

“Where would you refuse to automate?”

Anything irreversible I can’t undo in five minutes and can’t explain to the person who owns the P&L — payouts, production deploys, customer-visible publish — until there’s policy and a human on the path. I’ll automate the draft. I won’t automate the commit.

“Why should we hire you instead of a strong Staff already here?”

If you need another person to ship tickets on an existing agent, you don’t. If you need someone who will put a deny path, HITL on a scary write, and an eval that can fail CI into the first 90 days — and get another team to reuse it — that’s the work I’ve already done. I’m happy to be leveled by your bar. I won’t audition as a prompt engineer with a Principal sticker.

“What have you owned here / on the current seat?” (chapter 4)

[Charter win — one sentence: deny path / HITL / eval gate / standard adopted.] Lucid is still the 0→1 proof. This seat is the inheritance proof. I won’t pad it if the charter slipped.

Competing offer (only if it’s real and written)

I’m prioritizing you because [specific]. I have another written package at roughly [band]. Can we close the gap on [base / equity / sign-on / level]? I’m not going to invent a deadline you can’t meet.

Bridge counter: only if I’d actually stay.

I’d stay if we date the Principal path, expand the AI ownership, and keep it in writing. A vibe promotion isn’t a reason to pause the search.

Competing processes (numbers offline)

• Run loops in parallel when I can. Starting a fresh process after offer #1 rarely fits the clock.
• Only leverage real written offers. Bluffing destroys reputations. Recruiters talk.
• No fake deadlines. Give them room to work a compensation committee.
• Don’t bluff a leave I won’t take.

Score with Accept / reject rubric.

Prefer north-star when ownership + total comp + leveling are clear.

Accept bridge upgrade only if: Principal path is dated, AI ownership expands, and I keep search rights — or I consciously pause the search.

Decline: lateral title inflation, demo theater, founding without equity/authority.

Offer PDFs and exact asks stay offline.

Exit

• Signed north-star / peer Principal offer or
• Bridge upgrade with written leveling plan and dates or
• Documented postmortems + 90-day correction plan (next page) — not another platform

Related

• Leveling and reject postmortem
• Referrals
• Signal artifact
• Overview