AI makes producing results cheaper every day.

Producing the right results is still the hard part.

That distinction matters because it is easy to get pulled into an AI conversation about cheaper production. Better prompts. Better tools. Better models. Better hacks.

All useful, up to a point.

But if the work is wrong, unreliable, unreviewed, unsuitable for purpose, or pointed at the wrong target, cheaper production is not leverage.

It is cheaper grief.

Give an agent a vague goal, too much context, broad permissions, or no review gate, and it will still produce something that looks finished. The memo will have structure. The code may run. The plan will have headings. The recommendation will sound confident.

That does not mean the work is good.

It means the risk is prettier.

So what actually decides whether AI hands you leverage or grief?

Here is the focusing principle — and if you’ve read my book, you’ll recognize it as the premise the whole Leadership Superpowers framework stands on: human factors are always the largest factors.

Not as a slogan. As a way to look at the work.

The same capabilities that make human teams effective also make AI-enabled work effective: judgment, direction, context, constraints, feedback, calibration, accountability, and review. If those are weak, a better model may hide the weakness for a while. It will not fix it.

In fact, AI often exposes the weakness faster.

That is the crucible. AI heats the system. It reveals what is already there. Clear thinking gets more leverage. Sloppy thinking gets more expensive. Strong review loops get faster. Weak review loops get overwhelmed.

The tool is new. The leadership test is not.

But the furnace has a new source of heat.

Agent factors are where the test shows up

When you delegate work to a person, you manage human factors: clarity, skill, motivation, attention, trust, feedback, and accountability.

When you delegate work to an AI agent, you still manage all of that. But now you also manage the non-human conditions that shape the work.

I call these agent factors.

The model. The instructions. The context. The retrieval. The tools. The permissions. The memory. The interface. The workflow. The review loop.

That may sound technical. It is delegation.

And delegation is a leadership superpower — literally. In my framework, Direction is the superpower of making everything come together to achieve the objective, built in part from the powers of Delegation and Discernment. Judgment decides what the agent is even aiming at. Acuity decides whether you truly see what came back.

If you can lead people well, you already have the muscles this work demands. Agents just gave those muscles a new place to train.

These factors decide what the agent can see, what it can miss, what it can touch, what it can cite, what it can change, and what it will confidently present as complete.

This is why “write a better prompt” is too small. Prompting is one steering mechanism inside a larger system.

A prompt library built on today’s hacks is a depreciating asset. Change the model, change the harness, change the tool, and yesterday’s clever prompt may become tomorrow’s weird failure mode.

A learning system compounds. If you are spending real budget on AI capability, spend it on what survives the next model release.

That is the move worth making.

And it starts with seeing one trap clearly.

Polished is not finished

The most dangerous AI failure usually does not look dangerous.

It looks done.

The citation points to the wrong paragraph. The summary skips the item that mattered. The code works today but will be miserable to maintain. The analysis is confident about a number nobody verified. The workflow quietly uses stale context. The agent finds something plausible and misses the decisive source.

Everyone is wringing their hands about hallucination. Fine. Hallucination is only part of the challenge.

The bigger problem is apparent completion: work that looks finished before the evidence, judgment, and review have caught up.

A strong leader does not merely inspect the final answer. A strong leader asks, “What produced this answer, and how does the system get better next time?”

The point is to accelerate impact, not merely output.

And that starts with one small loop.

Do this one thing

For an AI workflow that matters, do not start with seven policies, twelve checklists, and a new committee.

Start smaller.

Pick one workflow your team already uses or wants to use AI for. Something real: research, code review, client deliverables, content, operations, decision support.

Then write down two things before the next run:

1. What one thing would make the biggest difference in the quality or impact of this output?
2. What one thing would most destroy its value — especially if it failed quietly?

The first question forces purpose. Not activity. Not “make a draft.” What would make the result useful for the human or business decision it is supposed to support?

The second question forces risk into the open. What would be costly if wrong, missing, stale, misleading, or quietly off target? A fake citation? A missing constraint? A stale number? A security mistake?

If you know the Leadership Superpowers framework, you will recognize these two questions. The first is Judgment pointed at an agent instead of a person: purpose before activity. The second is Discernment moved upstream: naming the failure that matters before it can happen quietly.

Nothing about the framework expires when the teammate is not human.

After the run, compare the output to those two answers.

If it failed, do not only fix the output. Fix the loop.

Add the missing source. Change the review gate. Narrow the permission. Require the test. Update the prompt if that is the real issue, but do not assume it is.

The point is not to have a better prompt.

The point is to never have the same problem twice.

That is how you get a system that learns. Not magic. Just a disciplined loop that turns every miss into a stronger next run.

Start with the highest-stakes work. Repeat the loop. Expand as you learn.

That is how you get more value with less grief — more right results, faster, with more trust in the work.

Where this is going

This issue opens a thread.

In the issues ahead, I will take the Leadership Superpowers — Judgment, Direction, Acuity, and the other ten — one at a time into their AI applications: how each one helps you get more from AI and stay ahead as the tools change.

The AI applications will live here, in the newsletter.

The foundation lives in the book.

Leadership Superpowers is not a book about AI. It is the book about the thirteen capabilities this whole argument stands on — the ones that decide whether any delegation, to a person or an agent, produces leverage or grief.

Get the book at gplead.com/book Buy print direct and the ebook comes bundled — start reading tonight.

The tools will keep changing.

The superpowers compound.

Best,

—Gregor

P.S. Reply with one AI workflow you already use or want to use — research, code, content, client delivery, operations, decision support — and I’ll suggest the one question I would ask before trusting it.

Forwarded this email? Subscribe here for more.