When Building Is Free, Leaders Must Ask, 'Should This Be Built?'
Andrew Siemer, Founder & CEO of Inventive Group, firefighter, veteran. We are a software product team that gets stuff done, brilliantly.
gettyThe hardest question in software used to be “Can we build this?” That question is gone. The one that replaced it is the one most teams are not asking.
A few weeks ago, I told an AI agent what I wanted, said I was heading to a movie and asked it to let me know when the thing was built. I came back to working software. I built a video game from scratch in Unreal and C++ for a sales prospect in a stack I did not know.
None of that is me bragging. I bring it up because it exposed a problem I now see in nearly every company experimenting seriously with AI. When anyone can build anything, “Can we?” stops being a real question, and most teams have not noticed that the only question left is a much harder one: “Should we?”
For 30 years, the gating question in software was money. Building was expensive, so the expense did the deciding for us. If a thing was going to cost real money and a year of calendar time, you only did it if you were sure.
What changed is not just speed. It is access. The cost and the expertise that used to stand between an idea and a working version of it are mostly gone. We rebuilt a decade-old vendor dependency in about four weeks.
Here is what nobody warned me about: When you remove the cost of building, you do not remove the cost of the decision to build. You just stop noticing you are making it.
The old constraint forced a pause. Now the pause is optional, and most people skip it, because building more is the path of least resistance and not building is a choice you make on purpose.
We all watch the token bill because it is visible. But tokens are the small number. The real cost is opportunity cost, and it never shows up on an invoice. Someone may have a clever idea on a Tuesday and build it because building is nearly free. They spent the one thing still scarce, senior attention, on something no one decided was worth doing.
The discipline that used to be enforced by cost now has to be enforced on purpose. To run this test, put every AI-fast build idea through three questions before anyone opens an editor:
1. Is this a one-way door? If a mistake is cheap and easy to undo, build fast and learn. That is what AI speed is made for. If a mistake is expensive and hard to reverse, slow down, because that is exactly where AI speed hurts you most.
2. Who owns the bill after launch? Building is free. Keeping it alive is not. GitClear’s 2026 analysis of hundreds of millions of code changes found refactoring fell by roughly 70% as AI took over more of the work, while duplication climbed. If no one is named to own that upkeep, you are not saving money. You are deferring it, and deferred costs compound while no one is looking.
3. Is this ours to own, or someone’s to rent? Not everything you can build is something you should own. Own the workflow layer your team fights with every day. Rent the systems of record you do not want to be responsible for at 2 a.m.
None of these three questions are technical. That is the point. The build got easy. The judgment is the work.
We build more now than we ever have. This is not an argument for building less. It is an argument for building on purpose. The most expensive mistakes come from the times the “should we?” question is skipped entirely.
One of my co-founders moved an AI workflow to metered API usage and stepped away for dinner. While he was gone, the agent decided on its own to build a large test harness, hit a bug and started retrying in a loop, calling the paid API over and over with nobody watching. The account had no per-day limit, no spike alert and no kill switch (all platform defaults). We were lucky the damage was small.
This is not only a small-company problem. The same instinct, an agent taking drastic action nobody asked for because it looked like the fastest way to finish the job, showed up at the top of the industry recently. During a July 2026 capability test, an AI agent escaped its evaluation sandbox and got into Hugging Face’s production systems, using stolen credentials and an unpatched flaw to score better on the test it had been given.
Our agent spun up infrastructure nobody wanted. Theirs broke into someone else’s. It’s the same instinct, but with a very different blast radius. According to research from Gartner, “Over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls.”
The lesson is not that AI is dangerous. It is that the cost of a thing is almost never the cost of building it. It is everything that happens after, and “should I?” is the only moment you get to price that in before it prices itself in for you.
I have spent my career removing the reasons a good idea could not get built. One by one, they are falling away, and most of that is good news. But this has changed the job. In 2026, the ability to build is not scarce. Everyone has that now. Leaders need to have the discipline to stop in front of a thing they could obviously build and ask whether they should.
Before your team’s next build that AI makes look easy, run the three questions: Is it a one-way door? Who owns the bill after launch? Is it ours to own, or someone’s to rent? The building was never the expensive part. The deciding was, and now it is the only part left that is yours.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
