Maximizing AI Value Without Minimizing Human Value
AI is lowering the cost of building software, but raising the value of judgment
Everyone is talking about AI transformation right now, but I’m not convinced most companies fully understand what they’re actually transforming into.
Over the last several months, I’ve been experimenting heavily with AI-assisted product development workflows across research, discovery, prototyping, customer validation, and strategic analysis. Not in a “look what ChatGPT can do” kind of way, but in a very practical enterprise product leadership way. The biggest realization for me has been that AI’s most immediate value is not autonomous replacement. It’s reducing the cost and friction required to learn.
A few weeks ago, I ran willingness-to-pay studies with nearly twenty target buyers across six different product concepts that, only a month earlier, existed as little more than rough concepts sitting in Jira Product Discovery. Historically, getting to that point would have required dedicated UX cycles, engineering conversations, stakeholder alignment, roadmap debates, and probably a small mountain of meetings. Instead, AI allowed those concepts to become tangible quickly enough that customers could react to them meaningfully before major investment occurred.
That changes the economics of product development more than I think most people realize.
For years, enterprise software teams have spent enormous amounts of time debating abstractions. Requirements documents. Priorities. Architecture assumptions. Workflow ideas. Everyone imagining a slightly different version of the future while pretending they’re aligned. What AI-assisted rapid prototyping changes is the ability to create believable futures early enough that customers, engineers, architects, UX teams, and executives can react to the same thing instead of reacting to their own internal interpretations of a document.
And the quality of feedback becomes dramatically better once something feels real.
Customers stop saying vague things like “that sounds interesting” and start saying things like “our compliance officer would never approve this workflow,” or “night shift nurses won’t use it this way,” or “this would save my managers hours every week.” That kind of operational feedback is incredibly valuable because it exposes hidden workflow realities before engineering hardens assumptions into code. The prototype itself becomes part research instrument, part alignment tool, and part strategic pressure test.
That’s why I increasingly believe AI’s biggest near-term enterprise advantage is not coding velocity. It’s learning velocity.
Ironically, I also think this increases the value of experienced operators rather than decreasing it. AI lowers the cost of manifestation, but it does not lower the cost of discernment. A weak operator with AI can generate endless polished confusion. A strong operator with AI can compress months of discovery, validation, and alignment into days because they know how to recognize weak assumptions, identify operational risk, and distinguish signal from noise. The leverage difference between those two scenarios is enormous and I suspect organizations are going to learn that lesson very unevenly over the next few years.
At the same time, there’s a second-order problem emerging that I don’t think enough companies are discussing yet. AI dramatically increases the production of plausible artifacts faster than human review capacity scales. Requirements documents, code, research summaries, architecture proposals, mockups, policies, presentations…suddenly everything can be generated almost instantly. The bottleneck is no longer creation. The bottleneck becomes whether anyone has the time, context, and judgment required to meaningfully validate what’s being produced.
That’s where things get dangerous. Because velocity without comprehension is not transformation. It’s operational debt with prettier screenshots.
This tension becomes especially important in regulated industries like healthcare, financial systems, compliance, and legal operations where “mostly correct” is sometimes indistinguishable from catastrophic. Human judgment does not disappear in those environments simply because generation becomes cheaper. In many ways, human accountability becomes even more important because organizations now have to govern not only people and systems, but increasingly autonomous or semi-autonomous cognitive workflows.
That realization is also why I think the SaaS business model itself is entering a deceptive stability phase.
Over the last several weeks, I pressure-tested this idea through long-form strategic discussions across ChatGPT, Claude, and Gemini. Interestingly, while the models framed the problem differently, they repeatedly converged around similar economic patterns. The conclusion we kept arriving at was that the near-term disruption is probably not “AI destroys SaaS.” It’s more likely that SaaS begins to stratify.
High-volume, seat-heavy, labor-interface workflows hollow out first as AI increasingly completes repetitive tasks end-to-end. But governance-heavy systems tied to operational authority, financial accountability, compliance, and regulated workflows likely remain resilient much longer. The “seat” survives, but not necessarily as a labor proxy. Instead, it becomes a governance construct. Organizations are no longer paying primarily for access to software. They’re paying for control authority inside increasingly agentic systems.
That distinction matters because many executive teams still seem to be treating AI as a feature when it’s rapidly becoming infrastructure.
And infrastructure changes margin structures.
It also changes talent structures. One of the things I worry about most is the growing trend of organizations reducing junior hiring because AI appears to increase senior throughput. That may look efficient in the short term, but expertise formation still requires exposure, repetition, mentorship, mistakes, and progressively earned responsibility. AI can accelerate learning, but it cannot fully replace lived operational experience. If companies stop building apprenticeship pipelines, they may discover several years from now that they optimized themselves into leadership shortages.
The organizations that ultimately benefit most from AI probably won’t be the ones that remove humans fastest. They’ll be the ones that learn fastest, validate fastest, reduce ambiguity earliest, and maintain strong judgment while everyone else drowns in generated noise. Because the real competitive advantage emerging right now is not simply the ability to produce more artifacts.
It’s the ability to understand which ones actually matter.