When Building Gets Easier, Deciding Gets Harder

Last week, I was on three different calls that all sounded strangely similar.

On one, a founder proudly demoed an AI feature their team had built in under a week. On another, a design lead confessed they hadn’t run research in months because “things are moving too fast.” And in a third, a PM walked me through a roadmap built almost entirely around what their new AI stack could do, not what customers had asked for.

No one was careless. No one was unserious. In fact, they were some of the most thoughtful people I know.

But there was a pattern: building has become dramatically easier. Deciding what deserves to be built has not.

That tension is everywhere right now. AI playbooks. Embedded builders. Analytics for agents. Stories of entire SaaS products spun up over a weekend. At the same time, conversations about skipped research, prediction amnesia, and quiet guilt about where we spend our time.

We’ve lowered the cost of production. We have not lowered the cost of judgment.

And that gap is starting to show.

The Production Boom Is Real

There’s no denying it: the mechanics of building have changed.

A 2024 GitHub survey found that over 90% of developers are using AI coding assistants in some form. McKinsey has reported productivity gains of 20–45% on certain software tasks when generative AI is used well. I’ve seen similar numbers in the teams I work with—prototype cycles that used to take two sprints now happen in a week.

One client recently built an AI-powered reporting assistant in ten days. Ten days. A year ago, that would have required a dedicated squad and a quarter of roadmap space.

The upside is obvious:

• Faster prototyping
• Lower experimentation costs
• Smaller teams doing more
• Less waiting for “perfect” specs

But here’s the part we don’t talk about enough: when production accelerates, constraint disappears. And constraint used to do a lot of quiet strategic work for us.

When engineering capacity was tight, you were forced to ask:

• Is this worth the tradeoff?
• What are we not building if we build this?
• Does this truly move the metric that matters?

Now, when something can be built in a weekend, the friction that once protected focus is gone.

And without that friction, many teams are drifting from strategy to possibility.

The Rise of Possibility-Driven Roadmaps

In the last month, I’ve reviewed five AI roadmaps. Four of them had a similar shape: a list of features enabled by new capabilities rather than anchored in validated problems.

You can hear it in the language:

• “We can embed an AI builder into the workflow.”
• “We can auto-generate insights.”
• “We can let agents take action on behalf of users.”

All impressive. All feasible.

But when I ask, “What decision does this help your customer make better?” the answers get fuzzier.

This is not a research failure. It’s a decision-making drift.

When tools make it easy to ship, the burden shifts upstream. The hard part becomes:

1. Choosing the right problem
2. Defining value precisely
3. Anticipating second-order effects (risk, misuse, confusion)
4. Designing for adoption, not just functionality

One team I worked with added an AI-generated summary feature to their dashboard. Usage was high—nearly 60% of users clicked it in the first week. Leadership celebrated.

But when we looked deeper, we saw something uncomfortable: users who relied heavily on summaries were 15% less likely to explore underlying data. Decision confidence went up. Decision quality did not.

The feature worked. The outcome degraded.

That’s the new complexity: AI doesn’t just add features. It reshapes behavior.

And behavior change is where product judgment lives.

Skipping Research Isn’t Laziness. It’s Overwhelm.

I’ve seen a lot of commentary lately about teams skipping research. The tone is often accusatory: “How could you not talk to users?”

But the conversations I’m having tell a more human story.

Teams aren’t skipping research because they don’t care. They’re skipping it because:

• The surface area of what they could build has exploded
• The pace of iteration feels relentless
• Leadership wants visible progress on AI initiatives
• The backlog keeps regenerating

When everything feels possible, everything feels urgent.

And research—especially generative research—feels slow by comparison.

But here’s the paradox: the cheaper it is to build, the more expensive it becomes to build the wrong thing.

In a pre-AI world, a misguided feature might cost three months. Now it might cost three weeks—but you can ship four misguided features in a quarter instead of one.

The volume multiplies the impact.

I worked with a SaaS company last year that added three AI features in rapid succession: auto-tagging, content rewriting, and predictive suggestions. Individually, each made sense. Together, they fundamentally changed the workflow.

Churn rose by 8% over two quarters.

Not because the features were bad. But because the product no longer felt stable. Users weren’t sure which parts were deterministic and which were probabilistic. Trust eroded quietly.

No single launch triggered alarm bells. The accumulation did.

Research isn’t about validating features anymore. It’s about understanding systemic effects.

Prediction Is the Missing Discipline

One trend that caught my attention this week was a tool that measures PM predictions instead of shipped output. It sounds niche. It’s not.

In my experience, one of the biggest judgment failures in modern product teams is this: we don’t write down what we expect to happen.

When shipping is fast, the feedback loop tightens. But if you haven’t articulated a clear hypothesis, speed just accelerates ambiguity.

Before launching any AI-driven feature, I now push teams to answer four questions explicitly:

1. What user behavior will change? (Be specific.)
2. What metric should move as a result?
3. What might degrade? (Trust? Skill? Clarity?)
4. How will we know if we were wrong?

This is less about governance and more about intellectual honesty.

A fascinating study from Tetlock’s work on forecasting showed that people who regularly make and score their predictions become significantly more accurate over time. The act of prediction sharpens judgment.

In product, prediction is how we keep speed from outrunning sense.

Without it, we end up in what I call retrospective rationalization mode—explaining whatever happened as if it was always intended.

With it, we create a culture where being wrong is data, not embarrassment.

The Identity Shift No One Is Naming

There’s another thread running through these conversations: a quiet identity crisis.

Founding CPOs coding 40% of their time. Designers building with AI agents. PMs debating whether they should be prompt engineers.

The guilt makes sense.

When tools make you individually powerful, it’s easy to drift into production because it’s tangible. You can point to what you built. It feels efficient. Useful.

But product leadership has never been about personal output. It’s about shared direction.

I’ve had to check myself on this.

There’s a deep satisfaction in spinning up a prototype on your own. But if that prototype bypasses the messy alignment conversations—about positioning, differentiation, tradeoffs—you may be optimizing for momentum at the expense of coherence.

The real work right now isn’t learning how to build with AI.

It’s learning how to:

• Say no to things that are easy to build
• Hold a clear thesis in a sea of capability
• Protect user trust when behavior becomes probabilistic
• Create focus when the toolset expands daily

In other words, the center of gravity is shifting from execution to judgment.

And judgment is slower. Quieter. Less demoable.

A More Boring, More Powerful Discipline

One founder I respect told me recently, “AI made us faster. So we made our process more boring.”

What he meant was this:

• Every AI feature requires a written hypothesis
• Every launch includes a pre-mortem on misuse and confusion
• Every experiment has a kill criterion
• No feature is considered successful based on novelty metrics alone

It’s not glamorous. There’s no Medium headline in it.

But their retention is steady. Their support tickets didn’t spike after their AI rollout. Customers describe the product as “smarter” but not “different in a scary way.”

That’s the bar.

In a world where anyone can bolt AI onto a workflow, differentiation won’t come from capability alone. It will come from restraint and clarity.

From knowing when not to automate.

From understanding which frictions are protective.

From designing experiences that feel coherent, not chaotic.

The future of product isn’t about who can build the most with AI.

It’s about who can decide the best.

And decision-making—real, thoughtful, accountable decision-making—has never been a tooling problem.

It’s a human one.

When building gets easier, deciding gets harder.

If we don’t invest accordingly, we’ll wake up surrounded by powerful products that no one fully trusts.

I’d rather be on the team that ships fewer things—clearly chosen—than the one that ships everything because it can.

That’s not anti-speed.

It’s pro-judgment.

And right now, judgment is the scarcest resource we have.