The Risk for Management Is Not AI, but False Competence

When “ChatGPT said so” becomes an argument from authority, the problem is not AI. It is the way we assess competence, decisions and accountability.

I have worked with AI for several years and, in the last two, it has become my only field of work. That is exactly why I start from a personal observation: from the first generative models, from GPT-2 to GPT-3, I too started producing better content than I would have made on my own.

Not because I had suddenly become more competent, but because I was using a tool capable of lifting my output.

If it was true for me, it is true for everyone. And this is where a theme is born that many managers are starting to see in their organizations.

Post-Competence.

I am not talking about a theoretical scenario. In recent months several customers have described a recurring dynamic to me. Managers with years of experience find themselves debating with junior colleagues who query ChatGPT and other LLMs during the meeting and bring the answer to the table as truth.

ChatGPT said so.

And so it seems true. It seems well-founded. It seems sufficient.

The point is that often it is not. Because an LLM can produce a formally impeccable answer even when the question is poorly posed, the context is undefined or incomplete, the assumptions are weak or the real problem is different from the one described in the prompt.

The risk is not that a junior person uses AI. On the contrary: using it well can speed up learning, analysis and synthesis skills. The risk arises when the AI's answer is treated as a shortcut to competence.

Management's new difficulty

Here management faces a new difficulty. In the past, a weak statement often looked weak in form too: poorly structured, poorly argued, unconvincing. Today that is no longer so.

A fragile thesis can arrive in a meeting with the tone, structure and language of a solid one. And this changes how organizations must evaluate contributions, people and decisions.

After Post-Truth, Post-Competence is emerging.

It is an even more delicate phenomenon for those who lead a company: it is no longer just hard to tell what is true, but who is truly understanding and solving the problem.

Well-written reports. Tidy slides. Coherent analyses. Quick answers. All formally correct. But how much of this reflects a real capacity for judgment and governance?

Better outputs do not mean real competence

Organizations have always evaluated people also on the basis of the outputs they produce. It was an implicit but fairly stable model: greater competence generated, on average, greater quality. Today that relationship has weakened.

AI makes it possible to generate high-level output even when understanding is partial. If management keeps judging mainly the formal quality of output, it risks overestimating competence, underestimating fragility and making decisions on apparently solid but substantially weak foundations.

It does not only concern the inside of the company

It concerns the whole chain: suppliers, consultants, advisors, controllers. A well-written proposal, an impeccable presentation or an apparently coherent analysis can create a perception of depth that does not always really exist.

This is the illusion of competence. And it is a concrete risk for:

• Quality of decisions.
• Risk management.
• Selection of partners and suppliers.
• Governance of AI initiatives.
• Assignment of responsibilities.
• Talent evaluation.

How evaluation criteria must change

The point is not to limit the use of AI. That would be a short-sighted answer. The point is to evolve the evaluation criteria. Attention needs to shift:

• From output to process.
• From the answer to the reasoning.
• From form to understanding of the context.
• From speed to accountability.
• From confidence in delivery to the ability to handle exceptions.

A good managerial question is no longer just: “What did you produce?”. But also: “How did you get there?”, “What assumptions did you make?”, “What does the model lack to answer well?”, “Under what conditions could this answer be wrong?”, “Which part comes from your experience and which from the tool?”

AI Co-Work does not mean governing processes

These questions will become even more important because a further misunderstanding is spreading today. Many think that by individually using AI Co-Work tools such as ChatGPT, Claude or others, it is already possible to manage entire business processes.

But in most cases it is not so. These tools, used individually, are becoming mostly an evolved substitute for Office: they write documents, prepare analyses, tidy data, build presentations, organize document storage on your own PC, look for insights in your own documents and help produce content faster.

In other words, they supercharge Word, Excel and PowerPoint. But supercharging personal tools does not mean governing business processes.

A process is not just an output. A process has roles, responsibilities, data, exceptions, controls, handovers between people, information systems, authorizations and metrics. And here the difference arises. Using an LLM as an individual co-worker improves personal productivity.

Agentic AI means operational control

Introducing Agentic AI instead means bringing AI into the real functioning of business processes.

When you introduce AI Agents into business processes, you are no longer just generating content. You are creating operational entities that access data, carry out activities, interact with systems, make guided decisions and produce outputs with real impact.

In that context, Post-Competence is no longer just about a discussion in a meeting. It is about control of the organization. This is why, in the work we are doing with TeraBrain by Linegon srl, the focus is not only on the model itself, but on the behavior of AI Agents:

• Roles.
• Rules.
• Access to systems.
• Traceability of actions.
• Quality of outputs.
• Consistency over time.
• Human supervision.

Post-Competence is not a cultural issue. It is a governance issue.

In the coming months it will make the difference between those who use AI to strengthen the organization and those who, without noticing, start losing control of their own decision criteria.

Have you already seen this dynamic in your organizations? People arriving at meetings with AI-generated answers and treating them as definitive evidence?

I believe it will be one of the most important managerial topics of the coming years.