Human-Centered AI

All of software, for all of computing history, has been an attempt to empower humans by making computational power accessible. The entire discipline of UX/UI exists to bridge the gap between humans and computing power, big data, algorithms. Every generation of technology—from command lines to graphical interfaces to mobile apps—has been about lowering the barrier between human intent and machine capability.

Generative AI is a quantum leap in that progression. Suddenly the access point to compute and data is natural language.

What that unlocks is profound. Siloed knowledge and specialized expertise have always been barriers to fully realizing human potential. You needed to learn to code, or master specialized tools, or spend years acquiring domain expertise just to access capability that already existed. Those barriers are falling. The question is no longer "do you have the technical skills to use this tool?" but "do you know what questions to ask?"

The media coverage of AI is obsessed with the wrong questions. Every article asks "When will AI replace workers?" Every vendor pitch promises headcount reduction. The public discourse is fixated on elimination rather than amplification.

But the wins I keep seeing aren't coming from replacement—they're coming from collaboration. The real value emerges when AI makes experts more effective, not when it tries to eliminate them entirely.

From my time at Microsoft, I watched the same pattern repeatedly: building an internal tool that experts could use effectively took X amount of effort. Making that same tool ready for general public consumption took 10-20X more work.

The internal tool could have quirks. It could require context. Experts knew the workarounds, understood the limitations, could interpret ambiguous outputs. But the public, fully-automated version? Every edge case had to be handled. Every possible misinterpretation prevented. Every failure mode addressed.

That pattern maps directly to AI. The effort to go from effective human-AI collaboration to full automation follows the same curve—roughly 90% of the total project effort lives in that last mile. But the collaborative version, the one where humans stay in the loop, delivers 60-70% of the potential value.

Think about what that means for resource allocation. In the time it takes to perfect one fully automated system, you could implement collaborative AI across ten or twenty different processes—each one making your people significantly more effective.

Full automation sounds appealing until you consider what happens when it fails. If humans aren't engaged—if they're not learning, not practicing judgment, not staying sharp—then the handoff back to human control becomes dangerous. Skills atrophy without use.

The better model: tools that enhance human capability while keeping people engaged. Not autopilot that sidelines the human until something breaks. Augmentation that makes humans better at what they do, not absent from the process entirely.

Accountability stays with people. The AI becomes an expert consultant sitting alongside you—but the judgment, the decisions, the responsibility remain human. What matters now is whether you're good at using that expertise, not whether you've personally memorized it.

This changes what we should value in people. Creativity and critical thinking matter more than encyclopedic knowledge. The ability to ask the right questions matters more than having all the answers cached in your head.

The future isn't AI replacing humans. It's AI making humans far more capable—and that future is available today, if we stop chasing full automation and start pursuing partnership.