Beyond the Copilot: Why "Systems of Intelligence" Require Precision Skills Data

The copilot era is giving way to something bigger

For three years, the default enterprise AI move was a copilot: a general-purpose model layered over email, documents, and chat. Copilots made individuals faster. They did not make organizations any smarter about their people.

That is starting to change. In September, The Josh Bersin Company released Galileo Jupiter, repositioning its HR assistant as an intelligence layer that sits behind enterprise AI platforms like Workday, Microsoft Copilot, and ServiceNow. Launch partner HiBob described the moment as a move from traditional systems of record to systems of intelligence.

Bersin's own framing is the one every CHRO should sit with. Enterprise AI, he argues, is no longer a question of capability. It is a question of trust, and the organizations that win will be the ones with "the most reliable intelligence behind those tools."

We agree. And we'd add one thing: in HR, the most important intelligence behind those tools is an accurate picture of what your people can actually do.

Why generic models fall short in HR

Bersin made the core argument last year: general-purpose models are trained on statistically averaged public data, which makes them generic by design. In HR, generic gets expensive fast. When his team benchmarked leading LLMs against 30 complex, real-world HR prompts, the models produced hallucinated or incorrect answers in more than half of the queries.

Domain expertise fixes part of that problem. A system grounded in decades of HR research knows how to reason about succession, redeployment, or org design.

But knowing how to reason about a decision is not the same as knowing the facts the decision depends on. Ask any system of intelligence, "Who on my team is ready to lead our agentic AI rollout?" It can only answer from the capability data it is given. In most enterprises, that data is pieced together from job titles, resumes, course completions, and self-ratings. It is inferred, not verified.

The skills data underneath is thinner than leaders think

Our 2026 State of Skills Intelligence Report [REPORT LANDING PAGE URL] surveyed 1,000 full-time employees at U.S. organizations with 5,000 or more people. The findings show how far AI adoption has outrun the evidence about who can actually use it well.

• Adoption is real. 67.8% of employees use AI tools multiple times a week, up from 39.9% in 2025. 58.3% have been offered AI-specific training.
• Skills data already drives decisions. 78.6% say their organization uses skills data, at least partially, for hiring or for team and project assignments.
• That data often gets people wrong. 42.5% say they've been passed over, held back, or mis-rated because someone misjudged what they could do.
• Self-assessment won't close the gap. 62% of employees say they understand their skill levels. Yet Workera platform data from more than 22,000 assessments shows 7 in 10 people misjudge their own ability.
• Much of the real growth is invisible. 46.5% are building skills on tools their employer never provided. Whatever they're learning, the company can't see it.
• Managers are short on signal, too. Only 26.6% receive regular coaching from a manager or executive, and 21.5% receive none at all.

Put those together and the picture gets uncomfortable. Organizations are feeding skills data into more talent decisions every year, and most of that data is proxy.

AI workforce planning is only as accurate as capability verification

A system of intelligence amplifies whatever you feed it. Give it inferred skills and it will produce confident recommendations built on unverified assumptions, at scale and at speed.

The redeployment plan looks rigorous. The succession slate looks data-driven. The AI readiness dashboard looks ready for the board. None of it is more accurate than the profile data underneath. Gartner research has found that only 33% of business leaders believe their current talent data is sufficient for informed decisions. Layering a more sophisticated model on top of that data doesn't raise the ceiling. It just hits the ceiling faster.

Three workforce planning questions break down without verified capability:

1. Can we execute the strategy? Tool access is not capability. High usage tells you people are trying AI, not that they can apply it to the work that matters.
2. Who should we move? Redeploying talent based on titles and course completions is guessing with better formatting.
3. Did our investment work? Without a measured baseline and a re-measure, there's no defensible way to show that upskilling dollars changed anything.

What a precision skills layer looks like

If systems of intelligence are the new front end of HR, verified capability data is the foundation they stand on. That foundation needs five properties.

• Verified, not inferred. Proficiency is measured directly through assessments built on Evidence-Centered Design, not reverse-engineered from keywords on a profile.
• Decision-grade. The evidence is granular and defensible enough to stand behind a hiring call, a promotion, or a board report.
• Continuous. AI skills shift in months, not years. An annual snapshot goes stale before the next planning cycle.
• Consent-first. In our survey, 41.1% of employees said they would opt into continuous skills measurement if they owned the data, and another 31.2% are undecided. That's 72.3% who aren't opposed, with one condition deciding which way they go.
• Connected. One trusted skills layer should feed every system that makes talent decisions: ATS, LMS, HRIS, and the AI agents now sitting on top of them.

There's a trust point here, too. 69.1% of employees believe a human manager would judge their skills more fairly than an AI tool. The goal isn't to replace that judgment. It's to give it evidence. Employees would rather be judged by a human. They would also rather be judged accurately.

Smarter is not the same as right

The next wave of HR AI will be judged on one thing: whether its recommendations hold up. Domain expertise makes systems of intelligence smarter. Verified capability data makes them right.

Before you plug another agent into your talent stack, ask two questions. What does it know about your people? And how does it know it?

Get the full data. Download the 2026 State of Skills Intelligence Report to see what 1,000 U.S. enterprise employees reveal about AI adoption, skills, and career mobility. 

Ready to see verified skills intelligence in action? Request a demo.

‍