AI and the Future of Work: From Production to Direction

AI is shifting the center of gravity of knowledge work from producing the answer to directing, judging, and standing behind it. Drafting, coding, analysis, and first-pass synthesis are cheaper than they have ever been. What remains scarce — and what will keep determining who advances and who doesn't — is the judgment to decide whether an output is right, the accountability to put a name on it, and the curiosity to keep understanding a system well enough to improve it.

From production to direction

For most of the history of knowledge work, seniority tracked with how much someone could personally produce. Junior employees wrote the first draft; senior employees wrote more, faster, and with more experience behind it. AI breaks that correlation. A junior analyst with a capable model can now generate a passable memo, model, or piece of code about as fast as someone twenty years into a career. Production is no longer the scarce resource, and an organization that keeps rewarding it as though it were will optimize for the wrong thing.

What doesn't get automated away is the decision about what to produce, what tradeoffs are acceptable, and whether a given output is good enough to act on. That work — direction — was always part of senior roles, but it was often mixed in with production, so the two were never cleanly separated. AI forces the separation. Faster drafts don't create faster organizations unless the direction layer — review, decisions, approvals — can absorb the new volume.

Human judgment and accountability don't scale the way output does

A model can produce a hundred plausible answers in the time it used to take a person to produce one. It cannot tell you, with any reliability, which of those hundred should be trusted in your specific context, with your specific customers, under your specific constraints. That is a judgment call, and judgment calls carry consequences that someone has to own.

This is where a common assumption breaks down: the belief that keeping "a human in the loop" automatically makes a process safe. A reviewer skimming a fourth AI-generated document in twenty minutes, under pressure to clear a queue, is not exercising judgment in any meaningful sense — they are rubber-stamping. Accountability requires enough time, context, and stake in the outcome to actually evaluate the work, not merely a human name attached to the sign-off. Organizations that increase AI-generated volume without giving reviewers more time, narrower scope, or better tools for verification are not adding a safeguard; they are adding theater. This has direct implications for how work gets evaluated and rewarded, which is why performance management needs to change alongside the work itself.

Systems thinking and curiosity are the durable skills

If producing an individual artifact stops being the differentiator, what does distinguish a strong contributor from a weak one? Two things tend to hold up: the ability to reason about a system rather than a task — understanding how a change in one place propagates elsewhere, where the real constraints are, and what will break — and genuine curiosity about why something works, not just whether it works this time.

These are hard to fake and hard to shortcut with a model. A person who has only ever assembled AI-generated components, without wrestling with why an approach failed or digging into a root cause, tends to plateau quickly: they can produce output but cannot reliably judge it, extend it into new situations, or explain it to someone else who will depend on it. Curiosity is what turns exposure to AI-assisted work into actual expertise rather than a permanent dependency on the tool.

The apprenticeship ladder is losing its lower rungs

Historically, junior employees built judgment by doing large amounts of unglamorous work — reconciling the spreadsheet, writing the boilerplate, drafting the first three versions of a document that a senior person would mark up. It was slow and inefficient by design: doing the grunt work was how someone learned to see the patterns a senior person sees instinctively. AI now does much of that grunt work faster and, often, competently. This is a genuine problem, not a minor inconvenience, because it removes the mechanism organizations have relied on for decades to grow senior judgment from junior labor.

Developing junior employees now requires deliberately manufacturing the struggle that used to occur naturally: assigning problems without the AI-assisted shortcut, requiring people to explain and defend outputs rather than just deliver them, and rotating juniors through the parts of a project where judgment, not speed, is being tested. Developing senior employees looks different too — their value increasingly lies in mentoring judgment rather than modeling production, and in getting comfortable reviewing work they did not personally draft line by line, which is its own underdeveloped skill.

What this asks of leaders

None of this resolves itself through tool adoption. It requires leaders to treat workforce development as part of the operating model, not a side program that HR runs independently of how work actually gets assigned and reviewed. That means redesigning career ladders around judgment and systems understanding rather than raw output, protecting time for junior staff to struggle productively even when it's slower in the short term, and being explicit with senior staff that reviewing and teaching are now core parts of their job, not interruptions to it.

It also means building the underlying capabilities that make good judgment possible at scale: redesigning how decisions get made and who is accountable for them, and measuring whether any of this is actually producing better outcomes rather than just more activity. A workforce that is "AI-ready" is not one where everyone has a license to a tool. It is one where people at every level understand the system well enough to know when to trust an AI-produced answer and when to distrust it — and leaders who invest in building that understanding now will have a workforce that compounds in capability rather than one that quietly stops learning.

For a closer look at where these ideas run into organizational reality, see how they play out inside the mechanics of running a project. This is the broader argument behind Beyond Doing — the case for shifting attention from producing more to directing better.