Average Output Is Getting Cheaper, Not the Average Person

Construction of the dome of Florence Cathedral began in 1420.

The span was so large that the timber scaffolding of the time could not support it. Filippo Brunelleschi did more than design the dome: he also developed animal-powered cranes and lifting mechanisms to raise stones and bricks to that height.

Those machines moved materials upward faster. But the crane’s strength is not what has kept the dome standing for six centuries.

It is the geometry of its nested shells, ribs, hidden chains, and herringbone brickwork. A more powerful crane working within the wrong geometry would merely place more bricks in the wrong position faster. Opera di Santa Maria del Fiore

We make a similar fundamental mistake when discussing AI today:

We confuse the crane’s capacity with the value of the structure it produces.

According to Ramp’s July 2026 data, the top 1% of companies by AI spending spent a median of $7,400 per employee, compared with just $11.95 for the median company: a gap of around 620 times.

A striking figure. But it measures spending intensity, not success, productivity, or economic value created. Spending 620 times more on AI does not mean making decisions 620 times better. Ramp AI Index

Cursor’s 2026 research points to similar concentration. The top 1% of users produce 46 times more AI-assisted lines of code and integrate 15 times more code changes than the median user.

That is a significant difference. But lines of code and business value are not the same. Cursor itself notes that code volume alone is not a perfect productivity measure. Cursor Developer Habits Report

The real issue is therefore not who uses more AI.

It is who can place that capacity within a structure capable of turning it into economic value.

Research offers two findings that initially seem contradictory.

In an NBER study of 5,179 customer service employees, AI increased productivity by an average of 14%. The gain reached 34% among the least experienced and lowest-performing workers, while its effect on experienced workers was limited.

In a bounded, repeatable task, AI was therefore able to narrow performance gaps by making the methods of strong performers available to a wider group. NBER – Generative AI at Work

Research by Harvard Business School and BCG involving 758 consultants shows the other side. On tasks within AI’s strengths, participants worked more than 25% faster, and output quality rose by over 40%.

But when the tool was used on a business problem beyond its capabilities, performance fell by 23% compared with the group without AI. Moreover, the similarity of generated ideas reduced the group’s diversity of thought by 41%. Harvard Business School AI Institute, BCG

AI can therefore both narrow and widen the gap.

It raises the performance floor in work that can be standardized. But where context, cause and effect, and responsibility for outcomes matter, it makes the difference between the right and wrong method more visible.

That is why the claim that the average person is now worthless does not seem right to me.

What is losing value is not the person, but average output that is easy to produce, detached from context, and indistinguishable from other output.

We already see signs of this in finance. Cash-flow forecasts, valuation tables, investment notes, budget variances, scenario comparisons, and management reports can be prepared much faster. In the near future, producing these outputs will not be a distinctive competitive advantage on its own.

But the quality of a financial decision depends on more than the speed of preparing a spreadsheet.

Which economic behavior generates revenue? Why is the collection period changing? Which balance-sheet items carry currency risk? Under what conditions will debt-service capacity come under strain? Is a variance temporary or structural? How long does management still have to intervene?

These are not merely separate cells to calculate. They are connected, like the relationships that carry the dome’s load.

A model built on faulty assumptions can work perfectly mathematically. AI can run it faster, explain it more smoothly, and present it with more impressive charts.

But speed does not turn flawed financial logic into sound logic. Sometimes it merely makes false certainty more persuasive.

This is where Ainsideme’s starting point takes shape:

Not producing more financial output, but turning corporate finance experience and a company’s economic reality into a functioning, traceable, reusable method.

Calculations can accelerate. Scenarios can multiply. Data can be processed faster. But why assumptions were chosen, under which conditions outcomes change, and which risks genuinely require decisions must remain visible. Actual outcomes must also feed the next assessment rather than remain outside the model.

Viewed this way, AI becomes leverage that expands a well-designed method’s capacity, rather than an answer machine replacing financial judgment.

Access to information is becoming cheaper. The capacity to produce text, code, analysis, and images is expanding rapidly. Yet knowing which questions are worth asking, recognizing hidden economic relationships, choosing sound assumptions, and taking responsibility for decisions remain scarce.

The most valuable companies of the future will therefore probably not be those spending the most on AI or producing the most output.

They will be those that understand which load must be carried, which relationships hold the structure together, and where accelerating is dangerous.

Because cranes are multiplying.

But the geometry of the dome is still difficult.