Share

Company Battlecards

On DeckKnow who to contact, what to say, and when to move

On Deck

Know who to contact, what to say, and when to move

SUBSCRIBE CONTACT US

Company Battlecards — Issue 6 | On Deck

Gate Diagnostic

The Crumple Zone Test

M.C. Elish named this in 2016: when control over a system is spread across machines and people, the human becomes the crumple zone, the part built to absorb impact. She was writing about aviation and nuclear operators. The mechanism is structural. Responsibility lands on the person with the least actual control.

You wrote the Decision Gate handoff. This is the failure mode sitting on the other side of it.

Every AI role description you read this quarter describes oversight. Almost none describe the authority underneath it. "Human in the loop." "Partner with product and engineering to define oversight patterns." "Own review and approval flows." Read those as claims about accountability until something proves otherwise.

Five conditions make a gate real. The person can see the system's actual state and where the output came from. They have time and training to evaluate it. Their authority to intervene is recognized outside their own reporting line. The system stops, reverses, or pauses when they say so. Using it costs them nothing.

Miss one and what you have is a signature line.

Run it on the posting first. Whatever the posting can't answer becomes a first-ten-minutes question.

Gate Diagnostic

The Crumple Zone Test

You wrote the Decision Gate handoff. This is the failure mode sitting on the other side of it.

Miss one and what you have is a signature line.

Run it on the posting first. Whatever the posting can't answer becomes a first-ten-minutes question.

TAKE NOTE

Five questions. Score the answer, not the enthusiasm.

State: "What does this role see before an output reaches a user or executes an action?" Weak: dashboards owned by another team, no trace or audit access.

Time and competence: "How long does a reviewer get, and who owns their training?" Weak: no number, no owner. Verification that expensive turns into rubber-stamping.

Authority: "Who can stop a release or roll back a deployed agent action, and is it written down?" Weak: "we'd escalate." Escalation is not authority.

System response: "If this role says stop, does the system stop, and who enforces that?" Weak: stopping requires an engineering ticket.

No penalty: "Has anyone used a stop or an override here? What happened to them?" Weak: nobody can recall an instance. Unused authority is decorative.

Bonus: "How do override decisions change the model or the design standard?" No feedback path means the gate prevents one bad outcome and teaches the system nothing.

PART ONE

Hold the Gate, or Draw the Screen After It

OPEN

Ninety days of executive appointments across your target list, screened for one structure: product and design under one leader, AI under another. The pattern is louder in conversation than in the record. The instances that survive verification point the wrong way: no fresh split produced public evidence that design gained authority over model launch. The companies handing design the gate never split at all. Five patterns below, ranked by how much each should change what you send this week. Names, tenure, contact paths, and the order to work them.

Hold the Gate, or Draw the Screen After It

PART TWO

Adoption Gets Funded. Control Gets Said Out Loud.

OPEN

Ten live senior design searches on the board. Nine could be checked for a time-bound success target attached to whoever gets hired, and none of the nine names one. The number that will actually judge this hire sits in a recruiter brief you will never be shown. There are only two numbers it can be: how fast AI moves through the organization, or what happens the first time it moves wrong. Buyers will describe both to you. One of them is on the performance review. Enterprise postings are the least honest about which. Here is how to hear it inside ten minutes, before you know anything at all about the person across from you.

Adoption Gets Funded. Control Gets Said Out Loud.

PART THREE

Vanta Let Engineering Take the AI Quality Story. This Posting Is the Buyback

OPEN

Nine months ago, the person at Vanta explaining how the company earns a user's trust in AI output was the VP Design. By June the same explanation ran under a different byline — SVP Engineering, more operational, with a named methodology and design nowhere in the loop. Then the VP Design left. What replaced her is a $365–513K Head of Design posting that never once says team building, sitting under a public commitment to ship ninety agentic workflows by December 31. Verdict: Act. Contact the CPO this week. And keep the five-to-forty-five story out of the first conversation.