When you can build a working prototype in an afternoon, the question is no longer can you make it — it’s should you build it, and what should you build.

Most founders and teams struggle with those questions, not because they lack conviction, but because they test that conviction too late. Resources get committed before customer needs are validated. Confirmation bias filters what gets heard. They often fall in love with solutions before fully understanding the problems they need to solve.

AI-accelerated development affects everyone — from solo founders building alone to small startup teams to larger software organizations at scaleups and enterprises. Whether you’re a team of one or one hundred, the challenge is the same: deciding what deserves to exist.

The CHAOS Report (Standish Group, 2020) indicates that more than two-thirds of software projects fail to meet their intended outcomes. While execution failures play a role, the problems often begin earlier — in the discovery phase. A focus on delivery velocity has replaced the judgment time needed to uncover what truly matters.

Co-Invention: Redesigning Discovery

I had spent six weeks conducting discovery and design on an AI-powered toolkit to accelerate product discovery when I came across Ajay Agrawal’s YouTube interview, “The AI Economist: The Skill You Need to Stay Employed in the Age of AI” (Full interview is 1 hour and 5 minutes). I was researching AI capabilities for the toolkit when his insight stopped me in my tracks. In my experience, software teams often skipped proper discovery, claiming they didn’t have time. I saw AI as the solution to a productivity problem: automate tasks, integrate workflows, and make discovery faster so software teams would have fewer excuses not to do it properly.

In their book Power and Prediction, Agrawal, Gans, and Goldfarb (2022) describe what happened when factories first adopted electricity. Manufacturers replaced steam engines with electric motors, but retained the same layout — one large motor, complex belts and pulleys, and machines clustered around a central power source. Productivity barely improved. The breakthrough came when they redesigned around electricity’s potential: machines could go anywhere, organized by workflow rather than power source.

This is what they call co-invention: the new technology is only part of the solution. The real value lies in inventing complementary systems, processes, and organizational structures that leverage the technology’s strengths.

Today we’re making the same mistake with AI — bolting it onto existing workflows, increasing speed without improving outcomes. An effective AI-human partnership requires co-inventing new systems that leverage the strengths of each. The processes that guide discovery need fundamental redesign, not around speed, but around judgment.

Speed without judgment means building the wrong things faster.

Prediction vs Judgment

Agrawal’s deeper insight reframed my thinking. In the human-AI partnership, AI makes predictions radically faster — filling in missing information, recognizing patterns, generating options, and forecasting outcomes. What remains scarce is judgment: the ability to interpret meaning, weigh trade-offs, and decide what’s worth pursuing.

Watch economist Ajay Agrawal explain this distinction(1 minute)

Decision-making is the interplay of prediction and judgment. AI handles what is likely to happen. Humans decide what should be done about it.

“AI handles what is likely to happen. Humans decide what should be done about it.”

In discovery, this plays out constantly. AI can analyze customer-interview transcripts and identify patterns across dozens of conversations — that’s prediction. But we must decide whether those patterns represent real problems worth solving or simply interesting noise — that’s judgment.

AI can forecast which features might increase engagement — that’s prediction. Only humans can decide how well they align with customers’ needs and the product’s vision, purpose and principles — that’s judgment.

Brandon Harwood (2025) describes this distinction in his framework for AI-powered product design, noting that specific tasks must remain human-centric due to the need for judgment, intuition, and emotional intelligence. What AI accelerates is synthesis; what humans provide is meaning-making.

Judgment develops by making decisions under uncertainty and owning the consequences. When you pursue an opportunity based on limited evidence, ship it, and learn it was the wrong bet, that failure can grow your judgment. When you kill a beloved idea because evidence doesn’t support it and that decision proves correct, that success strengthens your judgment. Learning comes from the whole cycle: decide → act → observe → reflect.

Discovery judgment is the practice of developing judgment about what deserves to exist. It’s not a single activity but a set of practices: identifying opportunities, talking to customers about their needs, mapping assumptions about what would make a solution valuable, testing those assumptions with minimal investment, and deciding when evidence is strong enough to commit resources.

Sinek’s (2009) Golden Circle, explored in Start with Why and his TED Talk “How Great Leaders Inspire Action”, provides a valuable lens for understanding why judgment remains a human trait. The Golden Circle shows that inspiring action starts with why — purpose and belief — then moves to how, and finally what you produce. This order is often reversed, starting with what or how to build.

AI can handle the how and even suggest the what. Only humans can determine the why, which makes it worthwhile to build. Judgment is the capacity to reason about purpose, meaning, and value — capabilities that belong uniquely to us.

Cross-Functional Discovery Judgment

If discovery judgment is scarce, everyone involved in bringing software to market needs to develop it. Discovery emerges from judgment calls across multiple perspectives — technical feasibility, user needs, market fit, operational constraints, and strategy.

Solo founders and small teams make all these judgment calls themselves, constantly switching between perspectives. You’re the product person AND the engineer AND the designer — often feeling overwhelmed by the need to think across domains that don’t come naturally.

Larger teams distribute these perspectives across various roles, including product managers, engineers, designers, user researchers, and customer success. When these perspectives stay siloed, teams consistently build the wrong things. The “that’s not my job” mindset creates blind spots that AI acceleration only amplifies.

Tutti Taygerly (2020) learned this at Facebook: true collaboration isn’t about stakeholders handing work to each other — it’s about establishing trust and reasoning together from the start. Whether developing multi-perspective judgment as a solo founder or learning to reason together as a team, the principle is the same: cross-functional discovery judgment determines whether you build the right things.

For solo builders, this means developing judgment across domains you might not naturally think in — and creating systematic practices to compensate for working alone. For teams, it means different perspectives combining to form a better collective judgment. Over time, this capability compounds — solo founders develop intuition across multiple domains, and teams learn to reason together effectively.

For organizations, siloed discovery wastes insight and resources. For individuals working alone, a lack of systematic practices wastes time and money. For professionals staying narrow in their function, they become replaceable.

Discovery as a Continuous Loop

Discovery isn’t a pre-build phase — it’s a continuous practice that runs alongside delivery in an ongoing learning loop. Whether working solo or in teams, you discover what to build, deliver it, learn from what happens, and feed those insights back into the next round of discovery.

“Discovery isn’t a pre-build phase — it’s a continuous practice running alongside delivery in an ongoing learning loop.”

Discovery requires double-loop learning: not only asking Did this feature succeed? but examining the assumptions and reasoning behind it. Single-loop learning adjusts what you do; double-loop learning improves how you think about what to do.

Matthew Godfrey (2020) explores this in his work on the critical role of discovery, noting that the best product teams recognize that discovery and delivery should run concurrently and continuously, rather than as separate, sequential phases. This dual-track approach enables teams to address problems as they arise throughout the product development process.

As builders move through the discovery–delivery–learning loop, they’re not just refining the product — they’re refining their judgment. Each cycle sharpens your sense of what matters, which assumptions to test first, and what evidence is sufficient to act on. For teams, when this capability becomes shared, organizations learn to reason together — and that capacity compounds. For solo founders, systematic practice creates the judgment that teams might develop through diverse perspectives.

Solo founders often think they can skip systematic practices because “I am the product person AND the engineer AND the designer.” But that’s why the loop matters more — you need systematic practices to compensate for not having diverse perspectives catching your blind spots in real-time.

Why This Matters Now

When spreadsheets arrived, accounting firms hired fewer accountants — but those who remained became more strategic and better compensated. Arithmetic specialists moved on; judgment-oriented accountants thrived.

Software builders face the same shift. The ratio of value to people is changing dramatically. Your value isn’t in what AI can generate — code, mock-ups, or transcripts. It’s in developing discovery judgment: framing the right problems, interpreting evidence, weighing trade-offs, and helping others do the same.

I’ve seen this pattern repeatedly throughout my career. Teams would spend months building technically impressive solutions — only to discover they’d solved problems no customer actually had. The breakthrough came when we started mapping assumptions systematically before building—the same talented people, but different discovery judgment practices.

The market is splitting: professionals who use AI merely to execute faster versus those who use AI to amplify their judgment. The latter will become increasingly valuable precisely because AI can’t automate judgment. As Teixeira and Braga (2025) document in The State of UX in 2025, algorithms and automated tools are taking over execution tasks, making human judgment about purpose and meaning increasingly valuable.

For solo founders and small teams, this shift is even more pronounced. You can now build a complete product yourself using AI — but your judgment quality determines whether that product solves a real problem. The democratization of building capability means your competitive advantage is 100% in judgment, not execution speed.

If this all feels like too much change, that’s natural. AI is rewriting how we work; uncertainty is real; change feels relentless. But you’re already making judgment calls every day — deciding priorities, interpreting feedback, choosing when to proceed or pause. It isn’t new work; it’s becoming deliberate about the work that matters most.

“Speed and judgment used to be a trade-off. Now they’re both required — and only one can be automated.”

Key Takeaways

• When AI makes building and delivering software easier, faster and more cost-effective, judgment becomes the scarce resource.
• Judgment develops through decisions under uncertainty and reflection on results.
• Cross-functional discovery judgment is the ability to interpret evidence and decide what deserves to be built and why — whether you’re building solo or as a team.
• Discovery judgment is a continuous, double-loop learning practice, not a phase.
• The software market is segmenting: those who focus on delivering faster versus those who compete on wisdom and strategic value.

References

Agrawal, A., Gans, J., & Goldfarb, A. (2022). Power and prediction: The disruptive economics of artificial intelligence. Harvard Business Review Press.

Agrawal, A. (2023, June 15). The AI economist: The skill you need to stay employed in the age of AI [Video]. YouTube. https://www.youtube.com/watch?v=UhfpHwcrx6c

Godfrey, M. (2020, June 12). The critical role of discovery in product development. UX Collective. https://uxdesign.cc/the-critical-role-of-discovery-in-product-development-6f50bf196722

Harwood, B. (2025, February 8). Collected consciousness: AI product design for empowering human creativity. UX Collective. https://uxdesign.cc/collected-consciousness-ai-product-design-for-empowering-human-creativity-dcc62100a7a4

Sinek, S. (2009). Start with why: How great leaders inspire everyone to take action. Portfolio.

Sinek, S. (2009, September). How great leaders inspire action [Video]. TED Conferences. https://www.ted.com/talks/simon_sinek_how_great_leaders_inspire_action

Standish Group. (2020). CHAOS Report 2020. The Standish Group International.

Taygerly, T. (2020, January 16). True collaboration with cross-functional peers (Engineering + PM + Design). UX Collective. https://uxdesign.cc/true-collaboration-with-cross-functional-peers-engineering-product-management-design-7be163369bb0

Teixeira, F., & Braga, C. (2025). The State of UX in 2025. UX Collective. https://trends.uxdesign.cc

About Gale Robins

I help software teams and solo founders develop discovery judgment — the ability to decide what’s worth building when AI makes building easier and faster. My approach combines methods such as Jobs-to-Be-Done, assumption mapping, and double-loop learning with evidence-based reasoning to make judgment development systematic rather than accidental.

Connect: www.linkedin.com/in/galerobins

When building software became easier with AI, deciding became harder was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.