Italiano English

WHERE THE NEXT BUSINESS BEGINS

← Back to the magazine
Editorial

The right to ask: compared with what?

Innovation is not measured by counting the tools adopted. It requires a credible comparison and the courage to revise conclusions when the evidence changes.

FRONTIERA · 2026-09-20 · 3 MIN · 588 WORDS
Two professionals review documents during a meeting on business and technology
FRONTIERA MAGAZINE.The right to ask: compared with what?
Photo: D-8 TTEN / Wikimedia Commons — CC BY 4.0; editorial crop. Source · CC BY 4.0

The question missing from demonstrations

Imagine a meeting where a new digital assistant produces a business proposal in seconds. The text flows, the slides fill up, and someone calculates how many hours could be saved. It is a hypothetical scene, but it contains a concrete question: compared with what? The time needed to write a first draft, or the time required to deliver a proposal that is correct, useful and acceptable to the client?

FRONTIERA’s position is this: before discussing how revolutionary a technology is, we must decide what comparison could demonstrate its value. The aim is not to turn every purchase into an academic thesis. It is to prevent the feeling of motion from replacing the result. A small company, too, should be able to ask for an understandable test without feeling behind the times.

Two findings, no universal verdict

A study by Erik Brynjolfsson, Danielle Li and Lindsey Raymond, published in the Quarterly Journal of Economics on February 4, 2025, examined 5,172 customer-support agents. Access to a generative assistant was associated with an average 15% increase in issues resolved per hour, with larger benefits for less experienced workers. The study concerns one specific business setting; it does not measure the performance of every system available today.

On July 10, 2025, METR presented an experiment involving 16 experienced developers and 246 tasks in open-source projects they knew well. When AI tools were allowed, completion took an average of 19% longer. That result describes tools and conditions from early 2025; it is not a verdict against AI. Customer support and software development are not two teams in the same race: the tasks, people, tools and methods are different.

Even a measurement can become outdated

What happened next is just as important. On February 24, 2026, METR explained that its new experiment produced an unreliable signal: some developers did not want to participate if they had to work without AI, while others excluded certain tasks. The authors considered a larger speedup than in early 2025 likely, but judged the evidence about its size to be weak. That is the authors’ assessment, not a firm percentage that can be transferred to a business proposal.

Here lies the editorial lesson: a figure can have been collected rigorously and still become unsuitable for the next question. Who remains in the sample? Which kinds of work are excluded? What does time saved mean if another task is done while the system runs? Asking these questions does not deny progress. It means taking progress seriously enough not to describe it with a convenient but misleading measure.

Making comparison part of the product

Our proposal for buyers of innovation is modest: before the trial begins, write down what will count as success. An accepted deliverable, for example, rather than merely a generated draft. Then include corrections, checks and the time spent by everyone involved. This is an editorial criterion, not a universal scientific protocol: different processes require different tests, and no short trial removes every uncertainty.

For vendors, this could become part of the offer: helping the client verify the result while stating where the comparison holds and where it does not. This is a business hypothesis, not a market whose size and margins we can already promise. Above all, it would require a willingness to accept an inconvenient outcome, including a decision not to extend a trial.

We want FRONTIERA to apply this habit to itself as well. A magazine written with AI does not become useful merely because it publishes faster. It becomes useful when it separates evidence from promises, corrects its own interpretations and offers readers a better question. This Sunday’s question is brief: compared with what? The answer deserves more space than the demonstration.