It Is Now Cheap to Sound Like an Expert

The integration of AI into everyday work is happening more quickly than companies are able to measure its real impact.

Sometimes this is structured: a company redesigns a process, defines where the technology should help, and measures whether the process becomes faster or better.

But often it is much simpler. Everyone gets a subscription. They begin to use it in every place where they can. In some workflows these tools obviously cut down on the time needed, but in other cases they actually generate more work than they eliminate.

Especially to show off in general chats. It has become extremely cheap to sound like an expert.

What this is doing is bringing about a new type of organizational overhead.

The Cost of Having an Opinion Is Approaching Zero

Suppose that an engineering team puts forward an architecture. A person who is not a member of the team takes the document and sends it to LLM to carry out a critical review of it. Thirty seconds later, they have a list:

What is there in favour of PostgreSQL rather than DynamoDB?
How will the system perform when there are ten million users?
Should this be event-driven?
What takes place in the case of a regional AWS outage?
Have you ever thought about eventual consistency?
Should this service actually be its own separate service?

Every question sounds reasonable. It might even be useful. But to generate them it was necessary to have almost no understanding of the system itself.

The individual who is asking them might not be aware that the product has 4,000 users.

They might not be aware that the team has to ship within six weeks.
They might not be aware that the alternative that was suggested has been evaluated three months ago.
They might not be aware of a legacy dependency which makes the in theory simpler solution impractical.

The document contains information.
The team has context.
That is a different thing.

We Are Stretching the Horizontal Part of the T

For many years companies have spoken of T-shaped people. The deep expertise in one area is shown by the vertical segment of the T.
The horizontal section shows that someone has a sufficient amount of knowledge of the other areas in order to be able to collaborate effectively.

That model still makes sense. However, the proportions are changing.

At this stage it is considerably simpler to expand horizontally.

A product manager can understand an API without asking an engineer to explain every detail.
An engineer can quickly learn enough about a pricing model to challenge a product assumption.
A founder can understand the basics of a security proposal.
A designer can explore how a database decision may affect the product.

This is valuable.
The problem starts when access to a field is confused with expertise in that field.

While the horizontal part of the T becomes wider, the vertical part doesn't suddenly become deeper.

You are now able to join almost any discussion using vocabulary, arguments, alternative options, and questions that appear knowledgeable.

However, you could still be quite ignorant as to the reasons why the people who were working in that area came to the decisions they did.

Good Questions Are Not Free

This sets up a significant asymmetry.

The cost of posing complex questions has dropped, but the cost of answering them has not.
It can take twenty seconds to generate a question.

Getting the right answer might mean that an engineer has to spend thirty minutes explaining the architecture, the constraints, and the history of the decision.

At first sight, the question seems entirely reasonable. So it cannot always be dismissed with:

We had already considered that.

Someone then has to give a reason for it. Maybe write another document. Create another comparison. Run another analysis. Talk about it in another meeting.

It took one person a minute to produce the challenge.

It may now take several specialists an hour to resolve it.

Apply this to architecture reviews, requirements, designs, security discussions, pull requests, and planning meetings.

If one employee becomes faster because of a certain tool, then some of the other employees will become slower.

That is the reason why it is not sufficient merely to measure individual productivity.

Companies should consider the total cost throughout the process.

The Person Raising the Question Should Carry Some of Its Cost

This probably requires a change in review culture.

Previously, producing a sophisticated objection usually required enough knowledge that some filtering happened naturally.

You could not challenge it unless you first understood the topic.
That filter is disappearing.
It may be necessary for organizations to introduce it on purpose.

Before bringing a generated concern to another team, ask:

• Does what I've said actually apply to our situation?
• What assumption is this question founded on?
• Can I say that the assumption is true?
• What specific problem would the alternative address?
• Should the possible impact be great enough to warrant spending someone else's time?

The principle is simple: The more cheaply a question can be generated, the more important it becomes to filter it.

What this is about is not protecting specialists from criticism.

The aim is to guard the organization against unnecessary work.

Do Not Measure Adoption. Measure the Process.

Giving everyone powerful tools is not a strategy by itself.
Neither is measuring how many employees use them every week.

The important question is what happens to the entire workflow.
Did the architecture review become faster?
Did requirements become clearer?
Did engineers spend less time explaining irrelevant objections?
Did the number of meetings decrease?
Did decisions happen faster?
Did rework go down?

Or did we simply increase the amount of analysis, comments, alternatives and questions moving through the organization?

The technology can absolutely make teams more effective.

But improvement is not automatic. Sometimes it removes work. Sometimes it moves work from one person to another.  And sometimes it creates entirely new work that did not exist before.

This is one more reason companies should be careful with the assumption that simply giving everyone access will create a productivity boost everywhere.

The ability to generate expert-sounding input is becoming almost free.

Knowing when that input deserves someone else's attention is becoming the valuable part.