Asher's Blog

Companies are beginning to train employees to use artificial intelligence, and much of that training concentrates on prompts. People are taught how to assign a role to the model, provide context, specify a format and ask for revisions. These techniques are useful, but they can create the false impression that AI competence is mainly the ability to phrase instructions cleverly.

A person who cannot define the problem, judge the evidence or recognise a weak answer will not become valuable because they learned a sophisticated prompt structure. They may simply produce more polished confusion.

Teach your team to think, not merely to prompt. Tool fluency will change as models and interfaces change. Judgment will remain valuable across every tool.

A prompt cannot repair an unclear objective

The quality of an AI result depends partly on the quality of the request, and the request depends on whether the employee understands the work.

If someone is asked to develop a growth strategy without knowing the target customer, business economics, current constraints or definition of growth, the model will fill the gaps with generic assumptions. The answer may be organised and persuasive while remaining irrelevant.

Thinking begins before the prompt. What decision are we trying to make? Which facts matter? What do we know, what are we assuming and what would change our conclusion? What consequence follows if the answer is wrong?

These questions give the tool a meaningful task and give the employee a standard against which to evaluate the output.

Tool fluency is temporary

Prompt techniques that appear advanced today may become unnecessary as models improve. Systems are already becoming better at interpreting ordinary language, retaining context and taking action across tools. A company that bases its training entirely on current interface tricks will have to retrain constantly.

Mental models travel better. An employee who understands unit economics can use any capable tool to analyse a product. A leader who understands root causes can interrogate a problem regardless of which model is available. A writer with taste can improve generated language without relying on one prompting formula.

The company should teach durable disciplines: problem definition, causal reasoning, evidence evaluation, estimation, ethical judgment and clear communication. AI then becomes an amplifier of capability rather than a substitute for it.

Require people to inspect the answer

One of the worst habits in AI-assisted work is accepting output because it sounds complete. Fluent language lowers suspicion, particularly when the answer confirms what the reader already hoped was true.

Employees should be trained to ask where a claim came from, what assumptions support it, whether the numbers reconcile and which important perspective may be missing. For high-consequence work, verification should be part of the process rather than a personal preference.

A useful practice is to require a short judgment note with important AI-assisted work. The employee explains what the system contributed, which parts were independently verified, what remains uncertain and why the final recommendation is theirs.

This prevents people from using the model as an invisible authority and reinforces human ownership.

Give the model disagreement, not obedience

AI is useful not only for generating an answer but for challenging one. Teams can ask the system to identify weaknesses, construct counterarguments, propose alternative explanations and show what evidence would disprove the current view.

This works best when the employee has already formed a preliminary judgment. If the model is always asked to think first, the user can become anchored to its framing. A healthier sequence is to define the problem, form an initial view, use AI to widen and test that view, then make a decision.

The goal is not to defeat the model in an argument. It is to make the organisation less vulnerable to its own blind spots.

Do not reward generated volume

AI makes it easy to produce long reports, numerous concepts and extensive plans. Managers may mistake this volume for productivity.

The employee who submits fifty pages has not necessarily thought more deeply than the employee who identifies the decisive issue in two pages. Performance should be evaluated through clarity, usefulness and outcome rather than the quantity of material produced.

This requires leaders to change how work is assigned. If the deliverable is “produce a report,” AI will produce reports. If the objective is “recommend whether we should enter this market, based on these criteria, and identify the evidence that would change your recommendation,” the employee must exercise judgment.

Build thinking into team rituals

Good thinking should not depend on a few naturally analytical employees. It can be reinforced through the way the company conducts reviews and meetings.

Ask people to state the problem before presenting the solution. Separate facts from assumptions. Require a definition of success. Examine the strongest alternative. After a result, compare what happened with what the team predicted and record what was learned.

These habits improve AI use because the team generates better context and notices weak output sooner. They also improve ordinary management.

Leaders should resist correcting every answer immediately. Ask questions that help the employee see the gap. If the founder always supplies the conclusion, the team may become skilled at operating tools while remaining dependent on the founder for judgment.

Judgment includes knowing when not to use AI

Not every task should be delegated to a model. Sensitive information may not belong in an external system. A difficult employee conversation may require personal presence rather than generated language. A high-stakes decision may need independent professional advice and accountable human approval.

Tool fluency includes understanding boundaries. Employees should know the company’s rules for data, permitted tools, review requirements and prohibited uses. They should also recognise when automation would make an interaction less humane.

The question is not “Can AI do this?” It is “Should AI participate, in what role and under whose responsibility?”

The thoughtful employee becomes more valuable

AI will reduce the value of people whose work consists mainly of transforming information from one format into another. It will increase the leverage of people who understand systems, customers and consequences.

The best employee will not be the person who memorises the most prompt templates. It will be the person who can take an ambiguous situation, define the real problem, direct intelligent tools towards it, reject attractive nonsense and stand behind the final decision.

Teach prompting, but place it inside a larger discipline. Teach people to ask better questions, recognise uncertainty, reason from evidence and communicate what they believe.

Tools will keep changing. A team that knows how to think will keep learning how to use them.