The house lights dim.
There is a brief rustle as people settle into their seats, programs folded, phones finally silenced. From backstage, you can see the front rows clearly, faces expectant, generous, ready to be taken on a journey.
There is a familiar mix of nerves and calm. Not because this will be perfect, but because it is ready; ready as it will ever be on opening night.
Ideas were thrown out without mercy. Scenes were cut. Lines rewritten. Awkward moments were inserted where they seemed necessary, then removed. Notes were taken. Notes were ignored. Changes piled up and were left behind.
Now there is only this version.
No more revisions. No more alternatives. No bargaining for one last change.
The stage manager gives the signal.
The curtain rises.
You’re on.
Taking the stage with AI
For any stage production, the moment just before the curtain rises on opening night is the culmination of weeks, months, sometimes years of work. And it is typically full of anticipation as well as anxiety.
After so long in rehearsal, the actual stage performance is the point where experimentation and practice end and real-time consequence begins. The audience just sees this one final version, not the discarded drafts, not the alternate endings, not the better idea that almost made it in, and so forth. All of that is washed out the instant the stage lights come up and the curtain rises.
It turns out working with generative AI is similar to this kind of performance lifecycle. AI tooling can help generate variations, test directions, and explore possibilities at a pace no human team could match. What it does not provide, however, is judgment; the ability to know what to cut, what to expand, and what to rewrite completely.
That work still belongs to humans with brains that are wired for decision-making as well as experimentation.
Three elements -- brainstorming, workshopping, and selecting -- sit at the heart of almost all creative work. That doesn’t just apply to the theater. It includes software architecture and programming, too. Especially when those activities involve generative AI tools.
So let’s look more closely at the creative lifecycle, and how generative AI has come to both enhance and undermine that process.
When generation became cheap and judgment became the work
For most of human history, generating options when engaged in creative problem-solving was the hard part. Ideas took time. Resources as well as alternatives were limited. Simply producing something new required effort, coordination, and often expense.
Under those conditions, judgment carried a different weight. Choosing was decisive because there was little sense that better options were waiting just out of reach.
This balance shaped how people learned to decide.
In the middle of the twentieth century, Herbert Simon described this reality as bounded rationality. Humans do not optimize in the abstract. They make decisions under real limits of time, attention, and information. What looks like compromise is often practical intelligence.
As Simon explained, choosing the option that feels right, what he called satisficing, is not about settling for less. It is about knowing when to stop. That instinct is wired into human cognition.
Generative AI disrupts this arrangement. Producing options is now cheap, fast, and effectively unlimited. When more alternatives are always available, stopping feels arbitrary, even irresponsible. Judgment, the thing that once freed us from worry, becomes a hurdle.
AI generators make it easy to collapse brainstorming and judgment into a single, continuous stream of text and images.
The problem is not that we have too many ideas. It is that we have lost the signals that tell us when enough is enough.
To see how this plays out in practice, it helps to look at the creative lifecycle one phase at a time, starting where abundance does the least harm.
Brainstorming. Generate without judging
Creative work has always made room for a phase where judgment is deliberately postponed.
Brainstorming exists to widen the field. Quantity matters more than quality. Ideas are allowed to be incomplete, awkward, even wrong. Evaluation is not just discouraged; it is deferred by design.
These rules were not invented to be polite. They exist because judgment, applied too early, collapses exploration.
Generative AI fits naturally into this phase. Used appropriately, it can produce variations quickly, suggest directions you might not have considered, and help fill the room with raw material. In this mode, it is an effective brainstorming partner.
The trouble begins when early outputs are treated as finished work. Polished language and confident tone make it tempting to switch into evaluation immediately, to start ranking and refining before the field has fully opened, before additional possibilities have been explored.
In creative practice, early sketches are not candidates. They are prompts for further thinking. When brainstorming works, nothing is chosen yet. The biggest mistake at this stage is deciding too soon.
Of course, exploration cannot stay open forever. At some point, the work has to stop expanding and start making sense of what it has produced.
Workshopping. Analyze without expanding
If brainstorming is about opening the field, workshopping is about holding it still.
In a workshop, the material already exists. Scenes are read. Songs are played. Drafts are shared. Designs are tested. The goal is not to add more content, but to understand what is already there. What works. What drags. What surprises you. What does not survive contact with an audience.
This phase depends on stability. The same material must be examined from multiple angles. Patterns only emerge when the ground stops shifting.
This is where many AI-assisted workflows fail. Humans struggle to apply critical thinking across a vast field of options. Our attention gets taxed, and we give up too soon.
If generation continues during analysis, insight never accumulates. Each critique invites another variation. Each weakness triggers another prompt. The work keeps moving, but understanding dissolves, like cotton candy in the stream.
Workshopping is slower than brainstorming and more demanding. It requires attention, memory, and restraint. This is where stylistic knowledge, experience, and judgment actually matter.
Without this phase, selection becomes guesswork. With it, choices become well-grounded.
Performing. Select and commit
Performance is where the work becomes visible.
A version is chosen and carried forward. Alternatives fall away. What remains is not necessarily the best possible option in some abstract sense, but the one you are willing to stand behind. The one the team agrees to. The one that is good enough for now.
This step is often misunderstood as optimization. It is not. It is commitment.
As George S. Patton once put it, a good plan executed today is better than a perfect plan next week.
In performance, there is no revision happening in the background. The audience encounters a single expression of the work. The discipline of selection gives that encounter its clarity and force.
Generative AI has no natural sense of this moment. It cannot decide. Worse, it will happily continue offering variations and refinements long after a decision should have been made.
Selection establishes consequence. It marks the transition from exploration to action. Without it, work never leaves the wings, or the developer’s workstation.
Insight prepares the ground, but it does not finish the job. Eventually, someone has to decide what actually goes forward.
What makes this moment so uncomfortable today is not indecision itself, but the sheer volume of alternatives pressing against it.
Why abundance makes decisions worse
It is tempting to believe that more options naturally lead to better outcomes. In practice, the opposite often happens.
Psychologist Barry Schwartz showed that expanding choice increases anxiety, regret, and dissatisfaction. When everything is possible, every decision carries the weight of what might have been chosen instead.
Under those conditions, judgment becomes distorted.
As Daniel Kahneman demonstrated, overload pushes people toward faster, shallower decisions. Instead of careful evaluation, we rely on shortcuts. The first option that seems acceptable wins, not because it is best, but because we are ready to be done.
Generative AI accelerates this pattern. Outputs arrive quickly, confidently phrased, and neatly packaged. Faced with an abundance of plausible answers, people often decide sooner, not later.
Too often, they decide too soon. The result is a paradox. More options lead to less thinking, not more.
If this sounds familiar, it is because we feel it every time we sit down to work. What helps is not more theory, but a simple way to keep the phases from collapsing.
The value of GAS is not the acronym itself, but what it triggers.
GAS as shorthand
In my own work, I rely on a simple reminder to avoid collapsing exploration, analysis, and decision-making into a single moment.
Generate
Analyze
Select
GAS is not a methodology. It is a reminder that raw output is fuel, not product. Fuel only matters if it is burned at the right time and in the right way.
The value of GAS is not the acronym itself, but what it triggers. It keeps brainstorming, workshopping, and performing from collapsing into a single, exhausting blur.
In my experience, keeping them separate is one of the most effective ways to improve our ability to improve.
Remembering the sequence is one thing. Protecting it is another.
Artificial constraints are how creative work survives abundance
As AI tools remove friction, we need to restore some of it intentionally.
This concern is not new. Movements that push back against speed and constant acceleration tend to emerge whenever tools outpace human judgment. In the early 2000s, Carl Honore gave this impulse a name in In Praise of Slow, arguing that slowness, applied deliberately, is not inefficiency but a form of care.
Alex Osborn imposed strict rules on brainstorming to prevent judgment from collapsing exploration. Donald Schön described reflective practice as a way professionals slow down action long enough to learn from it. Different domains, same instinct.
Creative disciplines have always understood that constraints are not a limitation on expression. They make expression possible. They help determine when to explore, when to listen, and when to act.
As deployed today, generative AI removes many of the natural constraints that guide thinking so the signal has to come from us.
Working with AI means restoring these boundaries deliberately, not because the tool demands it, but because good work does.
All of this effort exists for one reason: to make the moment of commitment feel earned rather than rushed.
Knowing when to step on stage
In the end, the confidence you feel as the curtain rises, or the service goes live, does not come from certainty. It comes from knowing the right work happened in the right order.
Ideas were given free rein. Critique was allowed space to do its job. A version was chosen and carried forward. By the time performance arrives, there is nothing left to debate.
The next time you sit down at an AI chat window to do some work, remember to think GAS, and remember that knowing when to stop is an acquired skill and it is part of the creative craft.
