AI Does Not Get Promoted

Most executive conversations about artificial intelligence turn on one question: how much of the work can the system do? It is a fair question, and it leads somewhere unhelpful. Work differs in kind, and the difference has a simple measure — not difficulty, not seniority. It is time. Specifically: how long someone has to live with a decision before anyone can tell whether it was any good.

A scheduler finds out by Friday. An operations lead who redesigns a process finds out in about eighteen months. A chief executive who enters a new market may wait seven years. A board deciding what the company is fundamentally for may never get a verdict inside the tenure of anyone in the room. Elliott Jaques named this measure more than sixty years ago and built a theory of organizations on it. He called it the time span of discretion.

Hold that measure in view and the question changes shape. It stops being how much can the system do and becomes: what happens to the system’s job as the wait for evidence gets longer?

The answer is that AI’s role does not get bigger as work gets more senior. It changes into something else. Across the climb from immediate work to institutional work, artificial intelligence does four different jobs — doer, designer, advisor, clarifier — and each is quieter than the one before. The system produces less and prepares more. The person produces less and owns more. AI does not climb the org chart. It changes what it is for on every floor.

Why the wait, not the model, sets the job

The instinct to treat AI’s cleverness as the deciding factor is understandable and wrong. What decides is not how good the output looks. It is whether you will find out in time to fix it.

Where the world answers back fast — the invoice reconciles or it does not, the code runs or it does not — the system can produce. As the wait stretches, that stops being true. Evidence arrives late, tangled with other causes, or after the decision has hardened. No improvement in the model shortens the wait. A system you cannot correct inside the window where correction still matters should not be producing the answer inside that window — however good the answer sounds.

The pattern underneath

Before people act well, they move through four disciplines: they make sense of what is happening; they work out what it means; they frame which problem is worth solving; and only then do they solve it. I call this the Progression of Meaningful Response, and it has a gradient. Near the front line, solving dominates. As the horizon stretches, framing matters more, then sense-making, and at the top — where an institution’s purpose is at issue — meaning-making becomes the whole job.

Set AI against that gradient and one rule falls out:

AI works one discipline below the one the job actually demands.

Where the job is to solve, AI solves. Where the job is to frame, AI cannot frame — it optimizes inside whatever frame you hand it — so it designs the arrangement that solves. Where the job is to make sense of a murky situation, AI can assemble the raw material from which sense gets made, so it advises. Where the job is to decide what something means, meaning requires standing to be affected by the outcome — so AI can only hold the record against which meaning gets tested. It clarifies.

Job one: AI as doer

Days to a year. Front-line and team-lead territory. The work asks a person: solve it.

At the base of the organization, the problem itself has already been settled further up. Here AI is a doer, and a good one. It drafts, sorts, reconciles, routes, transcribes, summarizes, and produces first versions at a volume no team can match.

Two obligations survive automation, and both belong to a person. The first is the standard. When a person does the work, part of the standard lives in their hands — the tacit sense that something is off. When a system does the work, the standard has to be written down, because the system will execute the version that was specified, not the version that was meant. The second is the promise. A customer who receives an answer does not receive it from a model. She receives it from your company.

The failure here is easy to miss because it is cheap. Review becomes a click. The exception queue gets worked by people who have learned the system is usually right — which is precisely when they stop reading. This is where judgment abdication, the condition I named in AI in the Org Chart, first enters a company.

The test: can you name the person who could state the standard the system is meeting, and would they recognize a breach of it?

Job two: AI as designer

One to five years. Operations and general-management territory. The work asks a person: choose which problem is worth solving.

A step up, the object of the work becomes the method — and then the system that produces it. Here AI stops making the work and starts making the arrangement that makes the work: the routing rule, the eligibility threshold, the ranking logic, the escalation trigger.

This is the most consequential and least governed of the four jobs, for one reason worth stating plainly. At this level, a setting is a policy. A routing rule decides whose problem gets attention first. A threshold decides who gets served and who waits. A ranking model decides what your company counts as merit. Those are policy choices with consequences running years out, and they arrive dressed as configuration.

AI is genuinely strong here. But it tunes toward the objective you handed it. It does not pick the objective, does not weigh what the objective leaves out, and cannot tell you the frame is wrong — because from inside a wrong frame, a wrong frame just looks like a hard problem. Choosing the frame is the human’s work. The obligation is to insist every system design be written out as the policy it actually is, in language a non-technical leader can approve or refuse, and then signed by someone senior enough to carry it. If nobody signed, nobody framed.

The characteristic failure is a small decision quietly making a large one. A team tunes cost-to-serve, and eighteen months later the company discovers it has changed which customers it is for.

The test: has the design been written as a plain-language policy, and has someone with the right seniority signed it?

Job three: AI as advisor

Five to ten years. Enterprise leadership territory. The work asks a person: read the situation before anyone can be sure.

At this altitude the work is to set direction and keep the enterprise viable. The wait for evidence now runs past the point where evidence can govern the decision. From here up, the result arrives too late to help and usually too tangled to teach.

So the job changes from producing to preparing. AI becomes an advisor. It widens the range of options, surfaces signals from outside your normal line of sight, models second- and third-order consequences, and — most valuably — argues the other side with real force.

Be clear about what that is worth and what it is not. It is worth a great deal, because leadership at this level fails through narrowness more often than through error. It is not worth a conclusion. The system cannot be held to the outcome, has no exposure to the consequence, and cannot be corrected by evidence in time for the correction to matter.

The danger is that none of this looks dangerous. The output is fluent, well-organized, confident. It has the grammar of a conclusion. A leadership team under time pressure will read a tidy recommendation about an unknowable future as though it were analysis of a knowable one. That is judgment abdication in its most expensive form: a human name still on the decision, the reasoning belonging to something that cannot answer for it.

The discipline is easy to say and hard to hold. Ask for the case, not the conclusion. Ask for the strongest argument against alongside it. And require the person deciding to explain, in their own words with the document closed, why the bet is worth making.

The test: can the decision-maker rebuild the reasoning without the output in front of them, and say what would change their mind?

Job four: AI as clarifier

Decades. Board and ownership territory. The work asks a person: decide what the institution is for.

At the top, the work is to guide the organization as a responsible actor in the world and to anticipate its place in systems that will outlast everyone now deciding. Here advice is not merely unreliable — it is the wrong category. A system asked what your institution should stand for can only hand back the values it absorbed, in fluent prose.

But there is a real job. Institutions at this horizon face a problem no single tenure can solve: staying coherent over time. Boards turn over. Executives leave. Commitments made in one decade get inherited by people who were not in the room. Drift is almost never a decision — it is an accumulation of small, defensible departures that nobody is positioned to see as a pattern.

A system can hold that record and put it back in front of the room. It can reconstruct what was promised, to whom, and on what reasoning. It can show the distance between what a company says it values and what its decisions have actually done, across years. It can point out which commitment is being quietly traded away, and in which meeting the trade began.

That is clarification: the humblest of the four jobs and the most demanding. The clarifier does not answer. It makes the question answerable. It hands the organization an honest account of itself, so the people with standing — directors, owners, stewards — can do the meaning-making only they can do, because only they can be held to it.

The test: does the system make the organization’s contradictions visible to people with the authority to resolve them?

The four jobs at a glance

Four jobs, not four stages

A progression like this reads easily as a maturity model — as though a company starts with doers and graduates to clarifiers. It is not. All three time horizons run at once, and so do all four jobs. A well-run enterprise operates every one simultaneously: systems doing work on the floor, systems shaping arrangements in the middle, systems widening the leadership team’s field of view, and systems holding the institutional record for the board.

What must not run together is the posture. The damage comes from using one job’s posture at another job’s horizon: promoting the system too far, when a doer-grade tool ends up carrying advisor-grade weight; demoting the problem, when the organization deliberates for six weeks over a decision whose evidence would have come back in nine days; or letting a small decision make a large one, when a short-horizon setting quietly makes a long-horizon commitment.

What you cannot delegate

Gillian Stamp’s Tripod of Work holds that a manager creates the conditions for good work through three things: tasking — saying what the outcome is and where the limits are; trusting — granting real judgment inside those limits; and tending — keeping the work relevant, supported, and connected to why it matters. Two of the three transfer cleanly to a system. You can and must task it — a job description that names what it may do, what it may never do, when it must stop, and what evidence has to travel with its output. You can and must tend it — check whether its assumptions still hold, whether the world it was built for still exists. Most organizations task adequately and tend almost not at all, which is why deployed systems drift.

The third does not transfer, and the reason matters. Trust, in Stamp’s sense, is given to someone who can carry the consequence of the judgment they have been granted. It is a relationship, not a permission setting. A system can be authorized. It cannot be trusted, because there is nobody there to answer.

What to do Monday

Short enough to run in one session, for every place AI now touches the work.

1. Ask how long the wait is. Not how fast the system responds — how long before anyone can know whether the judgment was sound.
2. Assign the job. Doer, designer, advisor, or clarifier. The wait decides it, not the vendor’s claims and not the team’s enthusiasm.
3. Name what stays human. The standard, the frame, the bet, or the meaning — and the person who owns it by name, not by function.
4. Apply the one test. Will evidence come back while you can still fix it? If yes, the system may produce. If no, it may only prepare.
5. Read your settings as policy. Every threshold, rule, and ranking is a decision about someone.
6. Restore the record. For every consequential decision the system took part in, can you reconstruct who judged, on what reasoning, and what would have changed their mind?

The point of all this

Management, leadership, and stewardship are practiced at every level of an organization and across every horizon. What changes with altitude is emphasis, not presence. The same is now true of artificial intelligence: it is present everywhere, and what changes is what it is for.

That reframing is the real executive task. The question is not how much of the work AI can take. It is what the work in front of you actually demands — and therefore what the system is permitted to be: the thing that does it, the thing that arranges it, the thing that informs it, or the thing that keeps you honest about it.

A system can move the work. It can draft, rank, route, arrange, and argue. What it cannot do is answer for what the work does to another person. The longer the horizon, the more the work consists of exactly that — and the more the human role becomes irreplaceable rather than less.

AI does not get promoted. It gets quieter, and we get more responsible.

A note on sources

The framework this article compresses is set out in The Capability Advantage, the first volume of the Built to Endure series: the Layers of Capability, and the Levels of Work, Mode of Thinking, and Domains of Competence as recognition lenses. That volume also introduces the Management Horizon and the Triad of Direction, which Keeping Time and Boards That Steward build out in full. The Progression of Meaningful Response and the term judgment abdication are from AI in the Org Chart. The design of borders, interfaces, and escalation behind the tasking discussion belongs to The Autonomy Paradox.

The Levels of Work ladder rests on Elliott Jaques’s Stratified Systems Theory and on Gillian Stamp’s development of it within the Bioss tradition, including the Mode of Thinking model and Career Path Appreciation. The bands are cited here, not reproduced, and the time spans track Jaques’s strata. The value-adding themes that name the levels — Quality, Service, Practice, Strategic Development, Strategic Intent, Corporate Citizenship, and Prescience — are Bioss’s, and are used here with that attribution. The Tripod of Work is Stamp’s. What is mine is the arrangement: the horizons, the progression, and the four roles of artificial intelligence set against them.

References

• Jaques, E. (1964). Time-span handbook: The use of time-span of discretion to measure the level of work in employment roles and to arrange an equitable payment structure. Heinemann.
• Maitlis, S., & Christianson, M. (2014). Sensemaking in organizations: Taking stock and moving forward. Academy of Management Annals, 8(1), 57–125.
• Ruiz, J. J. (2026). AI in the Org Chart: A Leadership Guide to Implementing AI Without Losing Human Judgment, Accountability, and Trust. Elavant Press.
• Ruiz, J. J. (2026). The Capability Advantage (Built to Endure, Book 1). Elavant Press.
• Stamp, G. (n.d.). The Tripod of Work. Bioss. https://www.bioss.com/gillian-stamp/the-tripod-of-work/
• Stamp, G., & Stamp, C. (1993). Wellbeing at work: Aligning purposes, people, strategies and structures. International Journal of Career Management, 5(3). https://doi.org/10.1108/09556219310038846
• Weick, K. E. (1995). Sensemaking in organizations. SAGE Publications.

---

Jose J. Ruiz is CEO and Managing Partner of Alder Koten and Chairman of Anker Bioss. This article compresses ideas he develops at length in The Capability Advantage and AI in the Org Chart, both published by Elavant Press and available on Amazon. To discuss how these four jobs map onto a decision your board or leadership team is holding, start a conversation.