The dangerous confidence of the 80% solution

Jason has a useful way of thinking about AI-assisted work: AI is remarkably good at getting us from zero to 80%.

The problem is what happens next.

There are plenty of situations where 80% really is enough, Jason acknowledged. Not every internal tool needs to be a masterpiece. Not every interface requires painstaking originality.

But something changes when 80% becomes the default rather than a deliberate choice.

AI tools are exceptionally good at producing plausible work because they have learned from enormous amounts of existing work. That makes them natural pattern machines. Give enough people access to similar models, however, and those patterns can start converging.

Jason described the potential result as a “sea of sameness.”

There is an upside to that sameness. Familiar patterns can make products more usable and accessible. Nobody needs an airline to reinvent the meaning of a checkout button just to prove its designers are creative.

But sameness eventually creates its own opportunity.

“That last 20%. That’s where the differentiation is,” Jason said.

Think of AI as a power tool in a woodworking shop. It can make the cut faster and straighter. It cannot decide whether you are building a beautiful table or an ugly one.

That still requires somebody who knows the difference.

Expertise changes what you ask

One of the more interesting ideas in the conversation came from Ranjitha’s experience as an educator.

She argued that AI tools are only as valuable as the mental models and conceptual frameworks we bring to them. The distinction between a novice and an expert isn't simply that the expert knows more facts. Expertise changes the questions you know to ask.

“Can you recognize when the AI is producing good quality output,” Ranjitha asked, “and do you know how to interrogate it to get to better outcomes?”

That word—interrogate—matters.

Much of the discussion around AI tools still focuses on generation: How do I get the model to make something?

The more consequential skill may be evaluation: Why is this the right answer? What assumptions produced it? What is missing? What happens when a real customer encounters it? What evidence would change my mind?

For product development, that distinction is enormous.

Generating 20 interface variations in minutes sounds impressive. Generating 20 variations without knowing what makes one better for customers merely creates a larger pile to sort through.

Output is not insight.

And speed is not judgment.

Taste may become a competitive advantage

Expertise, judgment, craft and taste can sound like different words for roughly the same thing. Ranjitha and Jason drew an important distinction.

Ranjitha described taste as a “personal expression of expertise,” something built not merely through accumulated knowledge but through accumulated experience.

Jason pushed the idea further.

Experience, judgment and even craft can often be tested against some standard. Taste is slipperier. It represents the culmination of experience into a point of view.

“That is the uniquely human aspect of it,” he said.

Jennifer offered an analogy from the performing arts. She had recently spoken with a former professional dancer about two performers who could be equally accomplished technically, yet one consistently gets picked for the leading roles. Why? 

“That person is pushing past the technique, past the expertise, and trying almost to fail creatively to see if it works,” Jennifer said.  

That’s a useful way to think about taste in the age of AI. AI can help us master and reproduce the patterns. The interesting work begins when humans decide when (and how) to break them.

This is where the conversation about AI and creativity gets more interesting than the familiar debate over whether machines can be creative.

The practical question for businesses isn't whether an AI model can generate a clever headline, an attractive interface or functional code. Clearly, it can.

The question is whether your organization has enough human expertise to know when the result should be accepted, rejected or pushed somewhere unexpected.

Taste is what makes someone look at a perfectly competent solution and say: Not yet.

Craft isn't about taking longer

There is a temptation to romanticize craft as the slow alternative to AI.

Jason offered a better definition.

Traditionally, he said, craft has often been associated with time, the idea that excellent work must take longer. But AI forces us to reconsider that relationship.

Instead of time, Jason suggested thinking about attention.

“It's the focused attention on something to really elevate the overall experience,” he said.

That distinction matters for anyone working in product design with AI.

If a designer can accomplish in two hours what once required two days, the answer isn't to stretch the work back to two days to prove it was sufficiently crafted. The opportunity is to reinvest some of that saved time into understanding the customer, testing assumptions, critiquing alternatives and polishing the details that matter.

AI-assisted design should give us more room for judgment, not an excuse to eliminate it.

The next generation needs critique, not just creation

There is another problem hiding beneath AI's productivity gains.

How do beginners become experts if AI lets them skip the painful parts?

Bad first drafts, failed prototypes and questionable decisions aren't pleasant. But they have traditionally served as tuition. You learn what good looks like partly by producing things that aren't good and understanding why.

Ranjitha worries that using AI naively could make people “lazier and less rigorous.”

But that doesn't mean removing AI from the classroom or workplace.

Jason described design teams using AI to create multiple variations and then deliberately critiquing them. Suddenly, a team that might once have had time to examine three possibilities can examine 20.

That's a much more compelling model for critical thinking with AI.

Don't use the machine to avoid the exercise.

Use it to create more material for the exercise.

For leaders developing junior talent, this suggests a simple shift: stop evaluating people solely on what they produce. Ask them to defend it.

Why this design?

Why this customer journey?

What did the AI get wrong?

What would you change?

What did you reject, and why?

Those questions reveal whether someone is developing judgment or simply becoming proficient at operating a tool.

AI changes the value of making things

For decades, education and many workplaces have disproportionately rewarded generation.

Write the code. Produce the design. Create the presentation. Deliver the analysis.

Ranjitha argues that AI should force us to give equal weight to another skill: critique.

She sees creativity operating on both sides of the equation. There is creativity in generating choices, but there is also creativity in evaluating them.

“I think it is equally important to develop skills for generating things, as well as critiquing things and reviewing things,” she said.

That may prove to be one of the defining changes in the future of work with AI.

When production was expensive, the ability to make something was scarce.

When production becomes cheap, discernment becomes scarce.

For businesses, that means the goal shouldn't be 10 times more designs, features, campaigns or code. More output only matters if it improves customer experiences, increases conversion, reduces rework, strengthens retention or otherwise creates measurable value.

The machine can help us explore more possibilities.

Humans still have to decide which possibilities matter.

Curiosity is still undefeated

To close the conversation, Jennifer turned personal: What would each person tell their 22-year-old self to learn deeply before touching AI?

Jason pointed toward music, art and literature; the interests his younger, more pragmatic self didn't always consider sufficiently useful.

Jennifer chose curiosity.

And Ranjitha returned to the ability to critique.

Taken together, those answers offer a useful blueprint for human-centered AI.

Learn the craft. Develop a point of view. Stay curious. Generate widely. Critique rigorously. Talk to customers. Test what you believe. Let AI accelerate the mechanical parts of the work so you can spend more attention on the parts requiring judgment, empathy and intent.

AI can make average work astonishingly easy to produce.

That makes settling for average a choice.

And perhaps the most valuable skill we can develop isn't learning how to make the machine produce more. It's learning when to look at something technically correct, beautifully polished and instantly generated and keep pushing anyway.

As Ranjitha put it: “Developing that ability to see something and understand why it’s good or bad is really important.”

Additional resources

• Ranjitha Kumar on LinkedIn
• Jason Giles on LinkedIn
• Jennifer Artabane on LinkedIn
• AI-related resources on UserTesting
• Webinar: The AI-native product loop: build, test, learn—without slowing down — Featuring Ranjitha Kumar, this on-demand session explores how teams can integrate real customer validation into AI-speed product development rather than treating research as a separate step.
• Guide: Human insight for the AI-driven product development process — A practical look at how AI is changing product development and why understanding trust, control, outcomes, and the human-AI relationship matters as teams build faster.
• Podcast: The AI design trap: Why shipping faster isn’t enough — Roger Wong and Jason Giles discuss why faster AI-assisted design doesn't eliminate the need for design fundamentals, human judgment, and feedback from real customers.
• Blog: When human judgement becomes the real design advantage — Jason Giles and other design leaders explore why judgment, curiosity, accountability, and customer understanding become more—not less—important as AI accelerates design work