When building is nearly free, the wrong thing gets expensive.
Cheap software sends three bills that never appear in the quote. Your team pays all of them.
Written by NickStrategy & Adoption Lead, Zipline
The cost of building software has collapsed. A workflow that needed weeks of development two years ago can be put together in an afternoon. Most people conclude that companies should therefore build more. I have come to the opposite view. When building is nearly free, deciding what not to build becomes the most valuable work in the company.
Cheaper always means more
In 1865 the economist William Stanley Jevons noticed something odd about steam engines. As they became more efficient, Britain did not burn less coal. It burned far more, because cheap power found a thousand new uses. Software is on the same curve. Every team can now have its own dashboard, its own automation and its own assistant, so every team gets them.
The numbers show where that leads. S&P Global Market Intelligence found that 42 percent of companies scrapped most of their AI initiatives in 2025, up from 17 percent a year earlier. Those projects were not too hard to build. They were easy to build, and that was the problem.
The three bills
A tool that is cheap to build is not cheap to own. I think of every new system as sending three bills that never appear in the quote.
The attention bill. Someone has to learn the tool, read what it produces and decide whether to trust it. In 2025 researchers at BetterUp Labs and Stanford's Social Media Lab named one form of this cost in Harvard Business Review: workslop, AI-generated work that looks finished and moves the real effort to whoever receives it. In their survey of 1,150 American employees, 40 percent had received some in the past month, and by their own estimate each instance cost them almost two hours.
The upkeep bill. Every automation rests on assumptions: a file format, a price list, the way one supplier writes an invoice. Assumptions change quietly. Somebody has to notice, and in most companies nobody has been told that they are that somebody.
The trust bill. This is the expensive one. Each tool that disappoints makes the next one harder to introduce. A team that has watched three pilots come and go will meet the fourth with scepticism, and they will be right to.
Why we keep adding anyway
If the bills are this predictable, why do companies keep building? Partly because adding is how people solve problems. In a 2021 paper in Nature, Gabrielle Adams, Leidy Klotz and colleagues showed that people asked to improve something reach for additions and systematically overlook taking things away, even when taking away is the better answer.
Management writers have warned about the result for a generation. Michael Hammer's 1990 article carried its argument in the title: "Reengineering Work: Don't Automate, Obliterate." Automate a process nobody should be running and you get the same bad process, faster.
Ajay Agrawal, Joshua Gans and Avi Goldfarb draw a related line in Power and Prediction. A point solution drops AI into one step of an existing workflow. A system solution redesigns the workflow around what AI makes possible. Point solutions are quick to build and easy to pile up. Most of the value sits in the system.
AI makes the habit worse, because the machines share it. Ask a model to improve a document and it will usually make it longer. Ask it to improve a process and you tend to get more steps. Cutting is still a human job.
Three questions before anything gets built
This is why our first promise to clients sounds odd coming from a company that builds AI systems: we build less. Before we build anything, we want answers to three questions.
- Whose hour does this give back, and did they ask for it? If the person doing the work would not choose this task to hand over, the tool will sit unused.
- What stops when this starts? A new system that removes nothing is an extra system. We look for the report, the handover or the approval step that can go.
- Who notices when it is wrong? Somebody with the knowledge to judge the output has to own it by name. If nobody qualifies, the task is not ready to automate.
Sometimes the honest answer is a simpler one: a better form, a fixed template, a rule in software you already own. When that is the case, we say so.
The scarce thing
For most of the history of software, the scarce thing was the ability to build. That scarcity shaped how companies budget, plan and hire. It is gone. What is scarce now is the judgement to look at ten things that could be built by Friday and pick the one that deserves to exist. I think of that as taste, and I no longer see taste as a creative luxury. It is an operating skill, and it is worth more every time building gets cheaper.
Sources and further reading.
The research referred to in this note, for anyone who wants to read it in full.
- William Stanley Jevons, The Coal Question (1865).
- S&P Global Market Intelligence, survey on enterprise AI initiatives (2025).
- Kate Niederhoffer, Jeffrey Hancock and co-authors, "AI-Generated 'Workslop' Is Destroying Productivity", Harvard Business Review (2025).
- Gabrielle Adams, Benjamin Converse, Andrew Hales and Leidy Klotz, "People systematically overlook subtractive changes", Nature (2021).
- Michael Hammer, "Reengineering Work: Don't Automate, Obliterate", Harvard Business Review (1990).
- Ajay Agrawal, Joshua Gans and Avi Goldfarb, Power and Prediction (2022).
More field notes
Co-existence is a team sport.
Most advice about working with AI is written for one person at a keyboard. A company is not one person.
4 min readAI changes fast. People don't.
Every company has an absorption rate: the speed at which its people can take in a new way of working. Trust sets it, not technology.
4 min readEveryone wants AI. Few ask what for.
AI is brilliant at some tasks and unreliable at others that look just as easy. A plan names the task.
4 min read