High Agency

Listen to how people describe their problems. Low agency speaks in passive voice: "the project got delayed," "leadership hasn't decided," "the tooling doesn't allow it." High agency speaks in moves: "I haven't unblocked it yet — here's my next attempt." Same facts, different operator. The single most reliable predictor of who becomes indispensable is which language they think in, because language here is strategy.

This is not positive thinking. Constraints are real. Agency is the practice of finding the part of the constraint that is yours to move — there almost always is one.

Practice. Write down your three biggest current blockers exactly as you would describe them to a friend. Rewrite each in moves: what is the next action only you can take? Take one today.

Most of the permission people wait for does not exist as a rule anywhere — it is a social guess about what might annoy someone. The high-agency calibration is simple: for reversible actions in your domain, act and inform. For expensive or irreversible ones, ask — fast, with a recommendation attached. "I plan to do X by Friday unless you object" gets more done than "should I do X?" ever has.

The companion skill is absorbing the occasional correction gracefully. People who act and adjust are forgiven; people who act and defend are not.

Practice. Identify one thing you've been waiting for permission to do. Classify it: reversible or not? If reversible, do it this week and inform. If not, send the "I plan to, unless you object" message today.

A task-owner does what was asked and stops. An outcome-owner asks what the task was for, notices when the task won't achieve it, and changes course. In an AI-saturated workplace, task ownership is precisely what got automated — outcome ownership is the remaining job. The shift is visible in small moments: reporting "I sent the email" versus "they haven't replied, so I'm calling tomorrow."

Pick your outcomes deliberately. Owning everything is owning nothing; the skill is holding a few outcomes so completely that nobody behind you needs to check.

Practice. For each of your current responsibilities, name the outcome it serves in one sentence. Find one place where completing the task would still fail the outcome — and fix the gap, not the task.

The first question about any decision is not "what's right?" but "what does it cost to be wrong?" Two-way doors — reversible calls — deserve speed: decide, observe, adjust. One-way doors — hires, public commitments, architecture you'll live with for years — deserve deliberation. Most organizational slowness comes from one error: treating two-way doors with one-way-door ceremony.

AI shifted the boundary. Many decisions that used to be one-way (build the thing, commit the team for a quarter) became two-way, because finding out is now cheap. Recalibrate accordingly.

Practice. List the five decisions currently pending around you. Classify each as one-way or two-way. For every two-way door, decide it this week by the cheapest test you can run.

Good deciders think in bets: what are the outcomes, roughly how likely is each, and what is each worth? The arithmetic is crude and the numbers are guesses — and it still beats the alternatives, because it forces the question "what would make this wrong?" into the open. A 30% chance of a big win can justify acting; a 95% chance of a small win can be a trap that costs you the year.

The high-agency twist: spend effort raising the value of the losing branches. A bet where even failure produces reusable assets — knowledge, code, relationships — is a much better bet than its probability suggests.

Practice. Take a real decision you face. Sketch the three most likely outcomes with rough probabilities and values. Then redesign the plan so the worst branch still leaves you something. Decide.

Writing a decision down — context, options, choice, reasons, what would change your mind — is the highest-leverage leadership habit per minute spent. It forces clarity you didn't know you lacked, lets others challenge the reasoning instead of the conclusion, and builds the track record that earns you bigger decisions. A page is enough. AI can draft the structure; only you can supply the judgment.

The memo also depersonalizes disagreement. People argue with a document very differently from how they argue with a person — more honestly, less politically.

Practice. Write a one-page memo for a decision you are about to make: context, options considered, your call, and the evidence that would reverse it. Share it with one person who might disagree.

Rank everything you do by output-per-hour-of-you. At the top: decisions, designs, relationships, and anything that runs without you afterward. At the bottom: work a machine or a process could do, done by hand, again. Most professionals — even senior ones — spend half their week in the bottom of their own ranking, because low-leverage work feels productive and high-leverage work feels like procrastinating.

The audit isn't about working more. It's about noticing that one hour spent building a reusable system or making a delayed decision routinely outproduces a week of conscientious task-doing.

Practice. Log your work hours for three days in 30-minute blocks. Score each block 1–10 on leverage. Eliminate, automate, or delegate the lowest three hours — permanently.

Treat AI delegation as a management skill, not a tool skill. The same rules apply as with a capable new hire: give the goal and the constraints, not keystroke instructions; demand artifacts you can verify; review early before errors compound; and keep a written record of what worked so the next delegation starts smarter. The managers who were good at briefing people are suddenly the best AI operators — that is not a coincidence.

Push past assistance into true delegation: not "help me write this," but "own this whole report; I'll review your draft Thursday." The leverage difference is 10x.

Practice. Choose one recurring deliverable you produce. Write a standing brief for it — goal, audience, quality bar, examples — and have AI produce the next one end-to-end. Review like a manager, not a co-author.

Machines took the tasks, so what you delegate to humans changes: you delegate ownership. That means handing someone an outcome with real authority and real stakes, then resisting the urge to take it back the first time their approach differs from yours. The test of good delegation is not "did they do it my way?" but "is the outcome safe in their hands, and are they bigger than they were?"

Under-delegation is the leader's most expensive vice. It caps the team at your personal bandwidth and teaches your best people to leave.

Practice. Pick the responsibility you keep doing because "it's faster to do it myself." Hand the outcome to a specific person this week, with a written brief and a scheduled review — and do not touch it in between.

The 2026 team is small, senior, and surrounded by machine capacity. Every member is a span-of-control multiplier: one person plus agents covers what used to be a function. This breaks the old management playbook — there is less coordination to do, fewer status meetings to run, and far more depends on each individual's judgment. The leader's job concentrates into three things: direction, quality bar, and unblocking.

The risky corollary: a small team amplifies whoever is in it, including the wrong hire. Selection and standards matter more than ever, not less.

Practice. Sketch your team (or a team you know) as it would look rebuilt today: who is essential, what the machines absorb, what the leader's three jobs become. Identify the single biggest gap between that sketch and reality.

When execution is fast, a vague goal does damage at machine speed: ten people and a hundred agents sprint in nine directions. Direction-setting is now the leader's highest-stakes output, and it must survive being paraphrased — by a teammate to another teammate, and by a teammate to a model. "Make onboarding better" fails that test. "A new user reaches their first success in under five minutes without human help" passes it.

Write direction with the same precision you'd demand of a spec: the outcome, the constraints, the non-goals. Then repeat it until you are bored of saying it — that is approximately when the team has heard it once.

Practice. Take your team's current goal as written. Ask two people (or two AI sessions) to independently paraphrase what it implies they should do this week. If the paraphrases diverge, rewrite the goal until they don't.

"The AI did it" is the new "the intern did it" — and it carries exactly as much weight, which is none. The rule that keeps AI-augmented teams trustworthy is simple: whoever ships it, owns it. Signing off on machine output means you staked your name on it. Leaders set this culture by example — visibly reviewing what they ship, visibly owning what goes wrong, never hiding behind the tooling.

Pair it with blameless mechanics: when AI-assisted work fails, the question is "what check was missing?" not "who trusted the machine?" Punishing trust teaches people to hide their workflows; fixing checks improves everyone's.

Practice. Draft a one-page working agreement for AI-augmented work: what review every shipped artifact requires, who owns what the machines produce, and how failures get analyzed. Propose it to your team.

Every yes is a budget allocation of the scarcest resource you have — focused attention. High-agency people are not the ones who say yes to everything; they are the ones whose few yeses actually happen. A good no is fast, honest about the reason, and where possible offers the smaller thing you can do. The slow maybe — agreeing vaguely and under-delivering — destroys more trust than a hundred clean nos.

AI abundance makes this harder, not easier: when everything is possible, the discipline of refusing most of it is the whole game.

Practice. Find one commitment you are currently failing slowly. Convert it this week into either a real yes (scheduled, resourced) or a clean no with an honest explanation.

Two failure modes kill teams: silent compliance (you saw the problem and said nothing) and permanent relitigation (you said it, lost, and keep saying it). The professional move is the third path — disagree at full strength before the decision, in writing, with your reasons; then, once it's made, execute as if the idea were yours. Your credibility on the next disagreement is built by how you carried this one.

Leaders earn this behavior by making dissent cheap before decisions and expensive after — and by conspicuously rewarding the person whose objection improved the call.

Practice. Identify a live decision you privately disagree with. Write your dissent in five sentences and deliver it to the decision-maker before the deadline. Whatever they decide, support it fully — and notice what that costs.

High agency without integrity is just effective selfishness, and everyone can smell it. The compounding asset behind every great career is being someone whose word settles things: estimates honest even when disappointing, credit given accurately, mistakes reported before they're discovered. In an environment where AI can fake competence, demonstrated integrity becomes the rarest signal and the thing people actually follow.

The tests never feel like tests. They feel like a small fudge that no one would notice. Decide your policy before the moment, because in the moment you will be tempted to negotiate.

Practice. Recall the last time you shaded the truth at work — an estimate, a status, a "yes, it's tested." Write down what honesty would have cost then, and what the shading risks now. Correct one live instance.

When machines handle the busywork, your remaining value is concentrated in a few hours a day of genuine, high-quality attention — and everything in the modern environment is engineered to fragment exactly those hours. Guarding them is not a wellness habit; it is the core operating decision of an AI-native career. Decide, on purpose, what your best two hours each day are spent on.

The audit question: of yesterday's attention, how much went to things only you can do? Most people are horrified by their honest answer — which is the point.

Practice. For one week, book your best two hours each day against your single most important outcome, before anything else gets scheduled. Treat the booking like a customer meeting.

The people who do great work for thirty years are not the ones who never push hard — they are the ones who treat intensity as a renewable resource with a maintenance schedule. Sprint when it matters, then actually recover: sleep, movement, time with people who don't care about your job. Chronic 80% effort produces worse work than alternating 100% and 50%, and it produces it from a person who is slowly going dim.

Watch for the AI-era trap: when tools make output feel infinite, ambition expands to fill all available life. The constraint was never the tools. It is you, and you are the asset to protect.

Practice. Map your last three months: where were the sprints, and where was the recovery? If you can't find the recovery, schedule a real off-switch this month — and observe what it does to the following sprint.

AI made starting effortless, which made finishing rarer and therefore more valuable. The graveyard of 80%-done projects is where leverage goes to die: all of the cost, none of the compounding. Finishing is a distinct skill — closing scope instead of opening it, doing the unglamorous last mile (docs, migration, the announcement, the handover), and calling a thing done out loud so it can start paying you back.

Momentum is psychological infrastructure. Each visible finish makes the next one easier, builds the team's belief, and builds your reputation as someone things can be safely given to.

Practice. Inventory your 80%-done projects. Kill half of them explicitly — written, announced, archived. Finish one completely within two weeks. Notice which felt harder.