Generative AI has carried out one thing unusual to the economics of data work: it has dramatically lowered the price of producing concepts.
Any moderately succesful skilled with a chatbot can now produce a dozen believable methods, memos, product ideas, or marketing plans earlier than lunch. In some instances, AI lowers the price of execution too—however not almost as far or as quick. Delivery even a type of concepts nonetheless takes weeks, months, or years.
The result’s already displaying up throughout workplaces: extra initiatives than groups can carry, extra instruments than anybody can study, and extra priorities than any cheap individual can maintain of their head. Leaders hold layering on new work as a result of the price of imagining new work has fallen near zero. However the price of truly doing it hasn’t.
This creates a brand new administration problem: in an AI-saturated office, the bottleneck is not concepts. It’s execution.
A cutting-edge genomics lab solved this drawback a few decade in the past—twice.
The Broad Institute’s lesson in doing much less to get extra carried out
The Broad Institute, an MIT-Harvard biomedical analysis heart, skilled one of many quickest price collapses in fashionable technological historical past. When the primary human genome was sequenced in 2003, it took greater than a decade and price roughly $3 billion. At the moment, sequencing a human genome can take hours and price underneath $200.
That collapse created apparent alternatives, but additionally two separate crises at Broad.
The primary was operational. As sequencing turned sooner, samples moved by the pipeline extra rapidly than downstream groups might course of them. Work piled up at bottlenecks. The lab turned so overloaded that technicians began shedding samples.
The repair was to maneuver from a “push” system—the place every stage sends work downstream as quick as potential—to a “pull” system, the place every stage solely takes on new work when it has capability.
Then got here a second disaster, one that appears loads just like the AI office drawback.
As soon as sequencing itself turned low-cost and routine, the Broad’s innovation workforce confronted an explosion of concepts. New initiatives have been began continuously. Few have been ever completed. As an MIT case research put it, the group was “shedding the know-how management place it had labored so exhausting to realize.”
The answer was the identical self-discipline utilized to concepts.
The workforce created a visible map—actually Submit-it notes on a wall—of each lively undertaking and tracked the place every sat within the growth funnel. The train made two issues apparent: some initiatives have been redundant, and there have been at the least twice as many underway because the workforce might realistically deal with.
They created a undertaking funnel on the wall, and added a “hopper” earlier than it—a holding space the place concepts waited till capability opened up within the funnel.
In two years, the workforce reduce lively initiatives by greater than half and elevated the variety of initiatives that truly received carried out.
Why leaders hold including work
The Broad’s repair appears apparent in hindsight. It not often occurs in follow as a result of people are biased towards addition.
A 2021 Nature research led by researchers on the College of Virginia discovered that when individuals are requested to enhance a design, doc, or course of, they systematically default to including somewhat than subtracting.
Within the office, that bias compounds.
A brand new software will get rolled out, however the previous ones keep.
A brand new precedence is introduced, however previous priorities aren’t retired.
Extra conferences. Extra dashboards. Longer technique decks.
Most organizational complexity is the sediment of individually cheap additions made with out subtraction.
AI accelerates this dramatically.
It’s now trivial to generate a seventeenth strategic precedence, a fourth product line, or a 3rd dashboard. The bottleneck is not creativeness. It’s the people being requested to execute.
What high-performing groups do in another way
The businesses adapting finest to this shift are making use of some model of the Broad’s self-discipline.
Make lively work seen
You’ll be able to’t handle what you’ll be able to’t see. Put each in-process initiative on one shared floor—a wall, a dashboard, or a single doc. Visibility forces triage.
Cease beginning and begin ending
In operations analysis, limiting work in progress is among the easiest methods to enhance throughput. New work waits till one thing else is completed.
Outline “carried out” earlier than you start
Earlier than a undertaking begins, outline success clearly.
Tony Fadell, who led the design of the iPod and co-founded Nest, instructed me his most vital recommendation to startup founders is to write down the press launch earlier than beginning the undertaking. It forces groups to make clear priorities and outline the aim line upfront.
None of that is about undertaking much less. It’s about truly ending the work that issues.
In an AI-saturated financial system, concepts have gotten a commodity. The benefit will go to organizations that may determine which concepts are price doing, and that are price ignoring.
Tailored from INSIDE THE BOX: How Constraints Make Us Higher, by David Epstein. Copyright © 2026 by David Epstein. Revealed by Riverhead Books, an imprint of Penguin Publishing Group, a division of Penguin Random Home LLC.
