how to stand out in the ai age

When polished answers become abundant, give people a reason to trust your judgment.

AI can write your article, build your presentation, generate your code, and prepare your interview answers.

It can also do those things for the person competing with you.

That is the part missing from much of the advice about becoming more productive with AI. An advantage available to everyone needs something else to make it yours.

The question is no longer simply how much you can produce.

Why should anyone choose your work?

Stephen Klein’s LinkedIn posts and Vivienne Ming’s work offer a useful starting point. Both ask us to consider what happens to human capability when technology makes answers easier to obtain.

My conclusion is practical: learn to use AI well, then build evidence that you can recognize a worthwhile problem, exercise judgment, and improve an outcome.

There is no guarantee of being irreplaceable. There are better and worse ways to become valuable.

The danger of becoming professionally interchangeable

In a LinkedIn post introducing Alice 2.0, Stephen Klein argues that generic AI can make professionals sound polished and interchangeable. He connects reputation to the way people think, what they believe, and the experience behind their voice. LinkedIn

There is a commercial pitch attached to that argument. We should examine it as an argument, rather than accept it as proof that his product delivers the solution.

Still, the underlying concern deserves attention.

An article can be readable and contribute nothing. A strategy can be comprehensive and avoid every difficult decision. A security assessment can list fifty controls without identifying the exposure most likely to hurt the business.

Fluency makes these weaknesses harder to notice.

Research by Anil Doshi and Oliver Hauser illustrates a related tension. In an experiment involving short stories, access to AI-generated ideas improved evaluations of individual creativity, particularly for participants with lower measured baseline creativity. But the resulting stories became more similar to one another. Science Advances

That finding concerns a particular writing task. It does not establish that AI makes all work homogeneous.

It does show why better individual output and greater collective sameness can coexist.

My inference is that professionals who rely on familiar AI answers will need to work harder to demonstrate a distinctive contribution. Sounding competent will be a weaker signal when competence is easy to simulate on the page.

Learn to investigate the question

In another post, “The Craft of Thinking,” Klein describes thinking as a capability developed through practice. His proposal for reflective AI turns the interaction around: technology should also question the person using it. Stephen Klein

That is useful advice for anyone whose work involves diagnosis.

Consider an executive asking for an AI strategy.

You could generate a roadmap covering governance, data readiness, skills, pilots, and scaling. It might be perfectly respectable.

But what is the executive trying to change?

Perhaps customers wait three days for a decision. Perhaps employees spend hours reconciling conflicting records. Perhaps the company has a demonstration looking for a business case.

Those situations require different responses.

In enterprise architecture and cybersecurity, the initial request often contains an untested diagnosis. “We need another platform” may conceal unclear ownership. “We need more monitoring” may conceal alerts nobody is responsible for acting on.

The contribution begins when you investigate that diagnosis.

Ask what is happening, who experiences the cost, and what evidence would distinguish the possible causes. Then decide whether AI belongs in the solution.

A person who can produce a roadmap is useful. A person who discovers that the proposed roadmap solves the wrong problem can save the organization a great deal of money.

Make AI challenge something you actually think

Vivienne Ming’s work gives this discussion a sharper edge.

In her discussion of Robot-Proof, she describes using a “nemesis” prompt: she writes a chapter, then asks AI to challenge it. Crucially, she evaluates the criticism rather than accepting it automatically. This Is Your Brain

That sequence matters.

You need a position before you can test your reasoning.

Before opening the chatbot, write down your recommendation, the evidence supporting it, and the assumption you are least confident about. Then ask AI to find weaknesses.

For example:

Examine this recommendation as a skeptical reviewer. Identify unsupported claims, plausible alternative explanations, and conditions under which it would fail. Suggest evidence that would help distinguish the alternatives.

This is a review instruction, not a truth machine. AI can produce convincing objections that are as weak as the original argument. Check them.

A Microsoft Research study surveyed 319 knowledge workers and collected 936 examples of AI use. Higher confidence in AI was associated with less reported critical thinking; greater confidence in one’s own capabilities was associated with more. Microsoft Research

The study measured self-reported practices. It did not prove that AI causes permanent cognitive decline.

It does give us a reason to examine our habits.

After using AI, can you explain the reasoning in your own words? Can you identify where you disagree? Can you defend the recommendation when someone changes an assumption?

If you cannot, the finished document may be concealing unfinished understanding.

Build expertise through contact with consequences

“Be curious.”

“Be creative.”

“Be more human.”

These phrases sound encouraging. They become useful only when translated into behavior.

Curiosity means investigating an anomaly after the first explanation sounds satisfactory. Creativity means developing an alternative that survives testing. Empathy means discovering how a change affects the people expected to live with it.

None of these capabilities makes domain knowledge optional.

To challenge an encryption design, you need to understand keys, trust boundaries, dependencies, and failure modes. To question an AI agent’s permissions, you need to understand what it can access, what it can execute, and how its actions will be attributed.

Research on the “jagged technological frontier” reinforces the need for discrimination. AI assistance improved performance on some consulting tasks and worsened it on a task outside the system’s capabilities. Organization Science

The lesson is to test where assistance helps within your actual workflow. A tool’s impressive performance on one task does not establish its reliability on the next.

Experience matters when it changes what you notice.

Twenty years of repeating assumptions offers little protection. Experience that teaches you to recognize missing evidence, brittle dependencies, and misleading success measures is more useful.

Keep putting that experience under pressure.

Give people evidence they can inspect

If someone claims to be an AI strategist, what should we examine?

A certificate establishes that they completed a defined learning requirement. A polished article establishes that they can publish a polished article.

Neither, by itself, establishes that they can improve a consequential decision.

Show the work that lets another person judge your contribution.

For a security architect, that might be an anonymized assessment explaining how an apparently well-controlled agent could still perform an unauthorized action.

For an enterprise architect, it might be a decision memo explaining why a smaller intervention was preferable to a platform replacement.

For a coach, it might be a carefully bounded account of an experiment: the behavior someone tried, what changed, and what remains uncertain.

Include the alternatives you rejected and the evidence that changed your mind. Protect confidential information and distinguish measured results from estimates.

A claim such as “This design is resilient” is difficult to assess.

A record showing which failure was tested, what broke, and how the recovery changed gives people something concrete to evaluate.

This is also where accountability becomes visible.

Name who owns the decision. Explain the remaining risk. Define the signal that would trigger a review.

Adding “human in the loop” to a diagram does little if that human lacks the time, information, or authority to intervene.

Make the responsibility real.

Run a small experiment before collecting another framework

Pick one problem you understand well enough to evaluate.

A recurring delay. A weak control. A decision people keep postponing.

Write a short diagnosis before using AI. Ask AI to challenge it and propose alternatives. Verify the important claims against original sources, system evidence, or the people doing the work.

Then test the smallest useful intervention.

Compare the result with the starting point. Did it reduce errors? Shorten a delay? Expose a faulty assumption? Help someone make a decision?

Publish a brief account if you can do so responsibly.

You may discover that your uncertainty required another conversation or a real test. Another afternoon of thinking would never have resolved it.

You may also discover that your elegant solution addresses a problem nobody considers important. That is useful evidence. Adjust.

The temptation in the AI age is to keep preparing: another tool, another course, another collection of prompts.

Preparation earns its keep when it changes what you can do.

Choose a problem. Investigate it. Use AI where it helps. Test your judgment against reality.

Then let people see what improved—and why they should trust your contribution.