5 AI Lessons From Q3 Every CHRO Should Remember
Training Signals: AI adoption training requests often reveal unclear priorities, workflows, and standards rather than simple tool knowledge gaps.
Shadow AI: Unapproved AI workflows expose unmet employee needs while creating ownership, documentation, and governance risks.
Oversight Evidence: Effective human oversight requires documented challenges, meaningful decision rights, and measurable overrides in high-impact workflows.
Coaching Boundaries: The same AI analytics can improve performance or enable surveillance, depending on access, user control, and retention.
Strategic Patience: Faster AI-enabled work does not justify immediate restructuring; leaders need evidence before changing roles, expectations, or capacity.
Our reporting over the last quarter kept returning to the same point: the hard part of AI adoption starts after the tool works.
Once people can use AI, leaders have to decide which problems are worth solving, what should be standardized, who should challenge what the technology produces and how quickly the organization should change around it.
With this in mind, here are five lessons from Q3 to keep in mind to carry into your 2027 planning.
Continue Reading for Free
Create a free account to finish this article, plus get ongoing access to timely insights and practical resources.
Have an account? Log In
1. Training Requests Usually Signal a Judgment Gap
At AI4 in August, Markus McKay-Fleisch, director of professional services AI enablement at Smartsheet, described finding more than 50 separate Claude Projects built by employees across finance, customer support, sales and HR to create on-brand materials.
The problem was that employees didn't know somebody else had already solved the same problem.
That's increasingly the problem hiding behind requests for more AI training. Managers may be asking for instruction when what employees really lack is clarity about which problems are worth solving, what good looks like or how their work should change.
The same limitation shows up in prompt libraries. A better prompt can make a task faster, but a redesigned workflow can change how the work gets done altogether.
AI enablement can't stop at teaching people how to use the tool. It has to help them decide where using it is worthwhile.
2. Unapproved AI Is Telling You Where the Problems Are
Analiese Brown, chief people officer at Campminder, calls the fragmentation created by uncontrolled AI adoption “chaos debt”.
It can look like progress at first. Employees build workarounds, automate tasks and share tools with colleagues. Over time, organizations accumulate workflows with unclear ownership, little documentation and dependencies nobody intended.
That creates an obvious governance problem, but it also creates useful information.
If someone in finance builds their own reconciliation process, or a recruiter bypasses an approved system with an AI workflow, ask what drove them around the official process in the first place.
Shadow AI doesn't just show where controls are weak. But it can show where existing tools and processes aren't meeting the needs of the people using them.
3. Human Oversight Needs Evidence
Most AI governance frameworks still put a human somewhere in the loop.
The more useful question is whether that person ever disagrees with the system.
One way to test whether AI decision rights are real is to look at consequential workflows such as candidate screening or performance flags and count the documented overrides.
If the number is zero, that's worth investigating.
Maybe the system really has been right every time. Or perhaps employees don't believe they have meaningful authority to challenge it, don't know when they should or aren't recording those disagreements when they happen.
Naming a human as accountable is easy. Building a system that allows that person can exercise judgment is harder.
4. The Same AI Can Coach or Surveil
AI can make parts of work visible that managers previously couldn't see.
At Cognite, managers received analysis of their own hiring interviews. One learned that 90% of his questions had been closed-ended. Another discovered he'd spent 85% of an interview talking instead of listening.
The information went back to the managers themselves so they could improve.
That distinction matters because coaching and surveillance can come from the same tool.
Who sees the output? Can the person being measured use it? How long does the record persist?
Those decisions increasingly sit inside notetakers, transcription products and analytics platforms. By the time an organization starts debating whether a tool feels invasive, some of the most consequential choices may already have been made in its settings.
5. Faster Isn't the Same as Further
AI creates pressure to compress timelines.
Build faster. Deploy faster. Show ROI sooner. Turn saved time into measurable value.
But speed can also encourage leaders to make permanent decisions from temporary evidence.
That is why strategic patience matters more as AI accelerates work. A workflow getting faster doesn't immediately tell you what the job should become, whether the gain will persist or what employees should do with the capacity created.
AI tends to change work one task at a time. Some responsibilities disappear. Others become more important. New ones emerge.
Leaders need enough time to see that pattern before redesigning jobs, raising performance expectations or cutting capacity around an early productivity gain.
Planning season makes that especially important. Decisions being made now about budgets, roles and expectations will last longer than the pilots informing them.
What Q3 Left Us With
The AI conversation is getting more specific.
Adoption conversations are moving into the rearview. The questions now are whether employees are applying it to worthwhile problems, what unofficial adoption tells us, whether humans still exercise judgment and whether leaders can resist turning every early gain into an immediate mandate.
