Your AI Tool Is Only as Good as the Team Using It — And Most Teams Aren't Ready

The productivity promise of AI at work has a dirty secret baked into the data: according to BambooHR via HR Dive (2025), nearly half of all time employees spend on AI tasks is spent correcting AI output — not generating value from it. That single finding should reframe every board conversation about artificial intelligence (AI) return on investment. The question is not whether your organization has deployed AI tools; it's whether your teams are spending 50% of their AI hours as expensive editors of mediocre machine drafts.

This is the distribution problem no one is naming clearly enough. Give two engineering teams the same AI tool and outcomes diverge sharply: one ships faster with fewer defects, the other burns cycles validating confidently wrong output. GitLab (2025) documented exactly this split inside its own engineering organization, finding that some teams found real workflow gains almost immediately while others couldn't locate an entry point into AI-native ways of working at all. Same tools, same access, radically different results. The differentiating variable is not the software — it's what GitLab calls "AI fluency," the organizational capability to work effectively with AI systems rather than merely alongside them.

What exactly is AI fluency, and why does it differ from AI training?

AI fluency is not a certificate program or a prompt-engineering tutorial. It is the set of behavioral and cognitive habits that allow a team to calibrate when to trust AI output, when to override it, and how to structure workflows so that human judgment is applied where it actually matters. HR Executive (2025) frames this as "behavioral AI literacy" — the intentional practice of maintaining critical thinking even as AI handles more cognitive load. The risk HR leaders are not taking seriously enough: AI may not replace human thinking directly, but teams that do not practice active skepticism toward AI output will slowly choose to do less of it. Institutional judgment atrophies quietly, not all at once.

This is the mechanism worth understanding. Large language models and AI coding assistants generate plausible-sounding output by predicting statistically likely next tokens — they do not reason, verify, or flag uncertainty reliably. That means the value a team extracts is almost entirely a function of how well members can interrogate outputs, spot hallucinations, and know when the tool's confidence is inversely correlated with its accuracy. Building that skill is an organizational design problem, not a software procurement problem.

Does compensation solve the AI talent and fluency gap?

The instinct many executives reach for is compensation. If your AI-fluent people are rare and at risk, pay them more. The evidence says this is necessary but nowhere near sufficient. OpenAI in 2025 averaged roughly $1.5 million USD per employee in stock-based compensation — unprecedented for a pre-IPO company — yet still experienced high-profile defections to rivals including Meta, which was reportedly extending offers in the hundreds of millions of dollars. If OpenAI cannot retain top AI talent on compensation alone, your organization certainly cannot. What MIT Sloan Management Review's research surfaces is that retention at the highest skill levels is increasingly driven by access to interesting problems, autonomy, and the sense that the organization is genuinely building something — factors that compensation amplifies but does not replace.

The implication for operations leaders is uncomfortable: you cannot buy your way to an AI-fluent workforce, and you cannot train your way there with a one-time workshop. GitLab's internal playbook points toward a different model — systematic peer learning, deliberate sharing of what works across teams, and making fluency-building a continuous embedded practice rather than an episodic event. The firms that will create durable AI advantage are those where knowledge about effective AI use circulates laterally and fast, rather than residing in a few individuals who happen to have figured it out.

The decision hiding in your next planning cycle

Every organization right now is making a quiet bet. Either you treat AI fluency as an infrastructure problem — something to be solved by purchasing better tools or running another training cohort — or you treat it as a cultural and organizational design problem that requires sustained leadership attention. The BambooHR data suggests the default state, absent deliberate intervention, is that your workforce spends approximately half of its AI-related effort cleaning up the mess AI made. That is not a productivity gain; it is a productivity transfer from humans to humans, with an expensive middleman in between.

The strategic question for your next planning cycle: if nearly half of your employees' AI time is remediation work as of 2025, what organizational changes — not tool changes — would cut that figure in half by the end of next year?