Usman Shuja is the Chief Executive Officer at Bluebeam.getty​For the last several years, the AI conversation has centered on technical prowess: who can build the most sophisticated models, who can write the cleanest code and who can recruit the rarest machine-learning talent.That focus made sense when AI lived mostly in labs, prototypes and slide decks. It makes far less sense now.Despite unprecedented investment, enterprise AI is struggling to produce durable value. Nearly half of AI proofs of concept are scrapped before production, and abandonment rates of AI initiatives more than doubled between 2024 and 2025, from 17% to 42%, according to S&P Global Market Intelligence. MIT research (paywall) puts the failure rate for enterprise AI pilots even higher, with 95% failing to deliver measurable P&L impact. When leaders investigate why these efforts stall, the root cause is rarely a broken model.The technology generally works. What fails is everything around it.The real constraint in the AI era isn’t intelligence—it’s judgment, or the ability to apply intelligence inside real workflows shaped by context, risk and accountability.The Tension: Controlled Performance Vs. Operational RealityI’ve seen this tension play out the same way across industries. Most AI systems perform brilliantly in controlled environments. Given clean inputs and bounded problems, they do exactly what they’re designed to do. The trouble starts when those systems leave the lab and collide with the field.Real work doesn’t happen in pristine datasets. It happens amid partial information, shifting conditions, legacy systems and real consequences. In those settings, intelligence without context becomes brittle. AI can deliver recommendations that are technically correct and operationally wrong at the same time.That gap explains a large share of AI failures. Gartner research finds that 63% of organizations lack the data management practices AI requires.In physical-world industries like construction, infrastructure and healthcare, these failures may look like an AI tool generating an efficiency plan while missing a permitting constraint, optimizing a schedule without accounting for labor availability or offering what seems like a clinically sound recommendation that’s wrong for a specific patient.Two Forces Reshaping Who Benefits From AIThere are two forces at work. First, generative AI can boost productivity for novice and lower-skilled workers by more than 30%—and sometimes even more—by rapidly distributing best practices. The landmark 2023 NBER study found a 34% productivity improvement for novice workers across 5,000 customer support agents, driven by AI disseminating the tacit knowledge of top performers. Second, the same research shows that highly experienced professionals often see little to no productivity gain from the same tools.AI compresses the experience curve. It raises the floor. What it doesn't do is replace expertise, judgment or accountability. As execution becomes faster and more standardized, the advantage moves away from doing the work toward deciding what work should be done, and how.In other words, it shifts from technical execution to human judgment.The New Archetype: The Dual AthleteThis shift is producing a new professional archetype. I call it the dual athlete.A dual athlete isn’t an AI engineer and doesn’t need to be. They don’t train models or tune architectures. What they bring is something harder to develop: deep domain expertise paired with enough AI fluency to work productively alongside intelligent systems.There are four things that distinguish a dual athlete from everyone else:1. They know where automation helps and where it introduces risk.2. They know how to interrogate outputs, not just accept them.3. They can translate AI insight into operational action.4. They remain accountable for outcomes, not just outputs.That’s why the most valuable AI skill isn’t coding. It’s translation. You don’t need to be an AI expert to be AI-effective. But you do need to understand your domain well enough to direct the technology, challenge it and decide when not to use it.Three Shifts Leaders Need To MakeMany organizations still approach AI as a centralized technical initiative. They build centers of excellence, hire elite specialists and expect intelligence to trickle outward. That model breaks down quickly.​1. Decentralize intelligence. AI creates value when it’s placed directly in the hands of domain experts who understand context and consequence, not locked inside a central IT function.2. Invest in AI fluency, not just AI tools. That means building capability across the workforce, designing systems where every AI output has a human owner and rewarding judgment, not just speed.3. Treat human capability as the core asset. The organizations that struggle will be the ones that chase technical sophistication while neglecting the people who apply it.The Real Revolution Is HumanThe future belongs to dual athletes—professionals who combine deep domain expertise with AI fluency, who know when to trust the system and when to override it, and who remain accountable for outcomes no matter what the model recommends.The most valuable code of the next decade won’t be written in Python. It will be written in human judgment.Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?