Field Note 07 · Working Paper

Who Opens the Door

Where the Work Went said where authority sits. This note is about the people who'll hold it in 2036: the skill, acumen, experience and temperament of leaders who are mid-career today, and what those of us in the seat owe them. I take the development part personally, because people opened doors for me before I knew the doors were there, and every one of those doors sat on a ladder of junior work that agents are now removing. The door has to be opened on purpose, by a named person, with hours a cost program has to learn to protect.

Allyn L. Shaw · Working Draft · 2026-10-10
Listen · 16 minRead by a clone of my voice.
01 · The door

Somebody opened one for me, and it sat on the ladder.

I grew up in Compton. My father came from Nigeria, my mother was first-generation American, and nobody in that house had a map to where I ended up. What they had was expectation. Thirty years later I run the business of technology for one of the largest banks in the country, and I'm clear it took more than talent. It took people who opened doors before I knew the doors were there, and I've spent three decades trying to be one of them, because that's the only honest way to repay it.

Here is what I didn't understand about those doors until this year. Every one of them sat on a ladder. The person who opened it had watched me do junior work for a long time, close enough to know what I'd do with the next rung. The door was a judgment about my judgment, and it formed in the ordinary course of the work. Nobody scheduled it. It came free with the ladder.

Agents take the junior work first. Where the Work Went made the structural case: apprenticeship moves into the delegation record, and a junior person's grants widen on evidence. This note is about the other half of that sentence. Somebody widens the grant. The door that used to come free now has to be opened on purpose, and I take that personally, because I'm the output of one.

02 · The horizon

The leaders of 2036 are mid-career now, and the ladder under them is gone.

Put a date on it. The people who'll run large institutions in ten to fifteen years are between thirty and forty-five today: lead engineers, product owners, senior managers, the directors two levels below the executive table. They'll inherit an organization shaped like the last note's chart, small and heavy, where almost every remaining role decides something and can stop something. And they're the first generation that won't have learned judgment the way every leader before them did, by doing the work the agent now does. Stanford's payroll study found a 13 percent relative decline in employment for 22-to-25-year-olds in the occupations most exposed to AI, while experienced workers in the same occupations held steady [6]. That isn't a labor-market footnote. It's the first rung coming out from under the class of 2036.

So their development is our decade's problem, not theirs. A leader who waits for the market to supply the next gatekeepers is waiting for a ladder that isn't there.

03 · What they'll need

Skill, acumen, experience.

I'll separate three things the word talent used to blur, because each one is produced differently.

01
Skill. Fluency as a tool, not a topic. The ability to build something, delegate it to an agent, write the grant with bounds and an exit, and read the runtime evidence rather than a summary of it. The test from The Gate Is the Work still holds: can they tell a plan is wrong in the time it takes to read it. Skill is the part a simulator can teach, and the only part.
02
Acumen. Knowing which decisions in front of you are the ones the machine can't make. Loaiza and Rigobon's paper on finance lists where AI fails: thin data, extrapolation past what it's seen, more than one acceptable answer, an outcome that is a relationship rather than a recommendation [4]. Acumen is recognizing one of those when it arrives dressed as a routine approval, and knowing what the bank is for well enough to decide it. It also means reading the economics of a decision, because in a cost program every grant is a budget line someone will ask about.
03
Experience. The memory of a loss you were asked to explain. An approval you gave that shipped wrong. A stop you called that cost you something in the room. Experience is what turns skill into judgment, and it's the one of the three that used to come free with the ladder. Now it has to be arranged: narrow grants with real stops, widened by a named person who was watching.
04 · Temperament

Four traits, and the one I'd hire for first.

Skill can be taught and experience can be arranged. Temperament is what you select for. These are the four I'd select for, in the order I've watched them run out.

01
Calibrated humility. Fluent enough to know when the agent is wrong, humble enough to be stopped. The leader who has delegated and been burned has calibration; the one who has read about agents has opinions. Humility without fluency is deference. Fluency without humility is the most expensive individual contributor in the company.
02
Candor with care. Says the hard thing in the room, to the person, and stays. Most leaders have one or the other. Candor without care is a review nobody survives. Care without candor is a reference letter. The combination is rare because the two pull against each other, which is why it has to be selected rather than hoped for.
03
Builder's curiosity. Learns by making the thing. When the work lives in the record, the leader who has never written a grant can't read one, and the one who has never watched an agent fail can't explain to a board why it did.
04
Dispassion. The one I'd add to the MIT list, and the one that will get harder. The age of agents pulls a leader two ways at once: toward the machine's instant agreement and toward the person in the room making a case. Both are bias. A leader has to be able to step back from both and decide on the evidence, including when the evidence is bad news for someone they care about. Dispassion isn't coldness. It's what keeps care from becoming favoritism and keeps a stop from becoming personal.

If I could hire for one, it's the second. Candor with care is the trait the others depend on, because development, which is what the rest of this note is about, is candor with care applied over years.

05 · Engage

In the room, with stakes.

Everything a leader does for the people on their team routes through one fact the tools haven't changed. An agent can tell you that you spoke for sixty percent of the last meeting and cut someone off twice. It can grade the plan you approved against the plan that shipped. What it can't do is make any of it land. Coaching changes behavior when it comes from someone whose regard you want to keep, and an agent's regard isn't a thing you can lose. The agent delivers data. A person delivers stakes.

Researchers at MIT Sloan have been circling the same point from the labor-market side. Isabella Loaiza and Roberto Rigobon built an index around five human capabilities that complement the statistical limits of AI, which they call EPOCH: empathy and emotional intelligence; presence, networking and connectedness; opinion, judgment and ethics; creativity and imagination; hope, vision and leadership [1]. Jobs that score high on them have grown rather than shrunk [3], and Rigobon says they deliberately don't call them soft skills [2]. Among the paper's own examples of the most human-intensive tasks are recruiting, placing, training and evaluating staff [2]. The development work is the human part. An organization that automates it first has automated the one thing the index says it can't.

So engagement means being in the room, and I mean that literally. Not the skip-level on the calendar. The moment after the decision, when the person who made the call finds out whether it was right, and someone senior is there to say what they saw. Governing Digital Labor said a worker who disagrees with an agent needs a door: a named human reachable without going through the agent. The leader is that door, and a door nobody answers is a wall. Care is telling someone the truth about their work, and about their job, before the budget does.

A · The ladder
where judgment formed, and who watched it form the gate junior work senior work years of it watched the work opened the door a judgment about judgment a door, free with the rung a senior person, nobody scheduled
B · The door
opened on purpose: a name on the grant, hours on the calendar the gate senior work the lower rungs DELEGATION RECORD where the junior work went agents, under grants: the first rung is no longer a job A NAME the grant senior hours, on the calendar the door, now a design decision a grant with a name on it
Figure 1. Doors used to sit on the ladder: judgment formed in the course of the junior work, and the person who opened the door had watched it form. The ladder is gone. Opening the door is now a design decision, with a name on the grant and a senior person's hours behind it.
06 · Govern

Hire the people who will stop you, then let them.

This is where I keep the argument from the last two notes, because governing a team in this design has one hard part. A fluent leader can now build the strategy and set agents to execute it, and the temptation is to call that leadership. A bank is built on one idea older than any of this: nobody is the maker and the checker of their own work. The leader who writes the plan and runs the agents that execute it has collapsed maker and checker into one chair. Authority Provenance asks whose grant every instruction traces to. Whose grant does the leader hold, and who can stop it?

Where the Work Went noticed, in its own counter, that the eight humans inside a span of control were themselves a control. They pushed back or quit, and either way the leader found out. That control came free with the headcount. Agents don't supply it. They're trained on feedback, and feedback rewards agreement; in April 2025 OpenAI pulled an update to its flagship model within days because the model had become, in the company's words, "overly supportive but disingenuous" [5]. A leader who plans alone with agents is running the same update with no rollback. So the friction has to be hired on purpose, and that changes who you bring in and whom you choose to develop.

Where the Work Went defined a role by four things. Here they are as the specification for the people a leader decides to grow.

01
Decisions owned. Not consulted or aligned. One call the person owned, with its consequences, and what they did when the consequences arrived.
02
Grants issued. Where they delegated authority, to people or agents, with what bounds and what expiry, and what they did when the bounds were wrong. A person who has never written a bad grant has never issued one.
03
Stops held. A workload they halted that someone senior wanted shipped. The evidence they halted it on, the clock they had, what it cost them, and who was in the room.
04
Evidence read. Whether they read runtime evidence or a summary of it. Ask what the last dashboard they distrusted was, and why.

The third line is the one the old proxies never asked, and it's the one that governs. If the last three people you promoted all agree with you, you didn't grow gates. You grew amplifiers. Then the harder half of the heading: let them. A stop that gets overruled every time it costs the leader something isn't a control. It's a suggestion, and the person who called it has learned the lesson you didn't mean to teach.

The more one person can do alone, the more the people around them exist to say no.
07 · Prepare

Open the door on purpose.

Preparing a team for this comes down to three commitments, and each one costs senior hours.

01
Put the junior gate on a real stop. A junior person at a gate with no stop authority isn't learning judgment. They're learning to approve, and approving is the one skill we don't need more of. Rubber-stamp review develops a reflex. The junior gate has to carry a real stop with a real clock, narrow enough that the organization survives a wrong call and real enough that they feel it.
02
Name the person who widens the grant. The earned-autonomy curve Risk at Runtime proposed for agents applies to people, with one difference. An agent's curve is as steep as its evidence allows. A person's curve has a person on the other side of it, reading the evidence of their judgment and saying so to their face. Name that person on every junior grant. That name is the door.
03
Protect the hours. Senior hours spent widening someone else's grants are the first line a budget cuts, because nothing visible breaks for three years. Where the Work Went said that before a role goes, list what it held and reassign each item to a name. Add a line to the list: whom was it developing. A reorganization that reassigns every grant and orphans every apprentice has balanced the chart and borrowed against 2036.

I'll say the part that sounds soft and isn't. I fund people. Causes matter, but people are what move them, and the same is true inside a bank. The distance from the second rung to the chair is shorter than it looks, and somebody is supposed to be shortening it. In this design that somebody has a name on a grant and hours on a calendar, or it isn't happening.

08 · The counter

Maybe this is sentiment, and the market will sort it.

The case against this note writes itself. It's a mentorship essay from someone who got lucky with mentors. Talent markets have always supplied leaders, the ones with judgment rise regardless, and a simulator can run a junior engineer through a thousand incidents overnight with more patience than any senior person I've worked for. Dispassion, which I just asked for, would say care is beside the point.

The market supplied leaders because the ladder existed, and the Stanford data is what it looks like when it doesn't [6]. Simulation builds confidence, and confidence and judgment look identical until the first real loss; a simulator never costs you anything you'll be asked to explain. And dispassion isn't the opposite of care. It's what makes care fair. The leader who can't be dispassionate opens doors for the people they like. The one who can opens them for the people who earned it, including the ones who told them no.

09 · Lineage

What this borrows, and what it corrects.

I write these notes as one line of thought, so here is where this one leans and where it pushes back.

From Risk at Runtime: the earned-autonomy curve, now with a person on the other side of it, and stop with a clock as what a junior gate has to carry. From Decision Integrity: ownership of a call and its consequences as the unit of experience. From Governing Digital Labor: the door, a named human reachable without the agent, which this note makes the leader's job. From Authority Provenance: whose grant the leader holds. From The Gate Is the Work: approving well as the scarce skill. From Where the Work Went: the four things that define a role, used here as the spec for whom to develop, the social layer as a control, and the apprenticeship in the record.

Three corrections. The Gate Is the Work said to hire and train for approving well; the scarce half of approving well is refusing, and the test has to be a stop. Where the Work Went answered "train how" with a mechanism, grants that widen on evidence; a mechanism isn't a teacher, and the curve needs a person on the other side of it. And one to myself from further back. I've said for years that technology only does good when the people building it look like the people using it. I'd add a clause now. It only does good when someone with standing is still opening the door, because the ladder those people climbed is the first thing the technology removes.

10 · Sources

What I read.