GenAI is collapsing the distance between an idea and working software. That changes more than productivity. It changes what technology leaders need to be good at.
For most of my career in technology, ideas were expensive. Not because ideas themselves were scarce, but because turning one into reality required commitment. A new application meant people, budget, architecture, infrastructure, security reviews, development cycles, testing, deployment, and eventually operations.
We spent decades trying to remove that friction. We virtualized infrastructure. We moved to cloud. We adopted APIs, CI/CD, infrastructure as code, managed services, and serverless architectures. Each generation shortened the distance between idea and realized value. And that was progress.
But there was something hidden inside all that friction:
It forced us to choose.
When an idea required a team, a budget, and six months of delivery capacity, someone eventually had to ask:
Is this really worth building?
GenAI is changing that equation. Today, an idea can become a prototype in hours. Code can be generated, interfaces created, infrastructure defined, tests written, documentation produced, and multiple approaches evaluated before a traditional project would have finished its first planning meeting. As agentic systems mature, that gap will shrink even further.
That is extraordinary. But it creates a new leadership problem.
The Bottleneck Is Moving
For years, technology organizations optimized for one thing: execution velocity. How quickly can we move from requirement to code? From code to production? From production to customer value? GenAI dramatically accelerates that cycle, but when execution becomes abundant, execution itself becomes less differentiating.
The question starts to change from:
Can we build it?
to:
Should we build it?
Those questions may look similar. They are not. One is primarily a question of capability. The other is a question of judgment. And I believe judgment is becoming one of the most important scarce resources in technology.
When “Why Not?” Becomes Dangerous
There is a natural temptation when the cost of creating something falls dramatically:
Why not build it?
Why not add the feature?
Why not create another internal application?
Why not automate that workflow?
Why not spin up an agent?
After all, it might only take a few hours. And this is where the economics become deceptive.
The cost of creating software may be collapsing. The cost of owning software is not. Every production system still creates a footprint.
Someone needs to secure it, understand its data, monitor it, and support it. Someone will eventually need to integrate it, change it, govern it, migrate it, or shut it down.
And perhaps most importantly, every new system consumes organizational attention. That attention is finite. We could therefore enter a strange new era in which organizations generate technology faster than they can absorb it.
Not because the technology is bad. Because building became easier than deciding.
The Cost of Proving an Idea Is Collapsing. The Cost of Owning the Wrong Idea Is Not.
This distinction matters because the answer is not to become more conservative. Quite the opposite. GenAI should allow us to experiment more aggressively than ever before. If an idea can be tested in a day instead of funded as a six-month project, test it.
Build the prototype.
Put it in front of users.
Challenge the assumptions.
Try three different approaches.
Throw two of them away.
Use AI to explore the idea itself, not just to implement it. That may be one of the greatest opportunities GenAI gives technology leaders: dramatically reducing the cost of being wrong.
But we need to distinguish between experimentation and commitment.
A prototype is a question. Production is a decision.
That distinction becomes increasingly important when prototypes are so easy to create that they begin to look like finished products. Something working is no longer enough evidence that something deserves to exist.
Spend Judgment Before You Spend Tokens
I originally thought about this problem in terms of tokens. Before we spend tokens generating code, perhaps we should spend some human intelligence deciding whether the problem deserves those tokens in the first place. But the more I thought about it, the less this was really about token economics.
Tokens will become cheaper, models will become more capable, and agents will get better at executing longer, more complex tasks.
The truly scarce resources are higher up the stack:
attention, context, judgment, and accountability.
AI can absolutely help us with all of them. We should use AI to challenge our assumptions.
Ask it to argue against our idea. Analyze customer feedback, research alternatives, model different scenarios, prototype competing solutions, and look for second-order consequences we may have missed.
In fact, one of the best uses of GenAI may be helping us decide not to build something. But there is an important difference between using AI to improve a decision and outsourcing responsibility for that decision.
AI can generate possibilities.
Leadership still has to provide priorities.
AI can analyze information.
Leadership still has to provide context.
AI can recommend.
Leadership still owns the consequence.
That responsibility does not disappear simply because the technology became more capable.
Maybe We Need a Different Definition of Productivity
There is another implication here that I think technology organizations will eventually have to confront. If AI allows one team to generate five times as much software, is that team five times more productive?
Not necessarily.
Software output has never been the same thing as business value. GenAI simply makes that distinction impossible to ignore. Lines of code were always a poor measure of productivity. Soon, even features delivered may become a questionable one. If the cost of generating another feature approaches zero, the important measurement becomes what happens after the feature exists.
Did customers adopt it?
Did it remove friction?
Did revenue change?
Did operating cost decline?
Did risk decrease?
Did we learn something important?
Or did we simply create more technology? The ability to produce more cannot become an excuse to think less.
From Execution Velocity to Decision Quality
None of this means execution speed stops mattering. It matters enormously. The ability to move from an idea to a real-world experiment in hours is transformational. But imagine what happens when that capability becomes broadly available.
Your competitors have it, startups have it, and your employees have it.
Eventually, practically everyone has access to extraordinary execution capability. What differentiates organizations then? I suspect part of the answer will be decision quality.
Which problems deserve attention?
Which experiments deserve another iteration?
Which prototypes deserve production?
Which successful products deserve more investment?
And perhaps most difficult of all: which things that we can build should we deliberately choose not to build?
Technology leadership has spent decades removing the constraints around execution. GenAI may finally bring us close to what we always asked for: the ability to turn almost any idea into working software incredibly quickly.
Now comes the uncomfortable part. We have to become much better at deciding which ideas deserve that capability. Because when building is no longer the constraint, building more is not the strategy.
Choosing better is.