Over the past month, I have been listening to people working across different industries, roles and levels of seniority.
These were informal conversations, primarily within my professional network. They were not formal research interviews, were not designed to produce a representative sample and do not establish findings about the organizations in which these individuals work.
Four recurring signals are worth following.
People described real benefits from using AI. It helped them develop faster starting points, structure unfamiliar work, produce analysis, writing and code more quickly, and explore ideas that may otherwise have taken longer to develop.
They also described uneven output, additional verification, unclear expectations and concern about how reliance on AI might affect their own thinking and the development of less experienced colleagues.
AI is already changing how work is produced and reviewed, and how people develop the knowledge and judgment the work requires. Much of that change is happening through individual choices rather than deliberate organizational design.
AI use is spreading faster than shared intent is forming
Some people are being encouraged to use AI. Others are experimenting because they see a personal or local advantage.
Guidance about where AI should contribute, what value it should create and how its output should be evaluated was often limited or unclear.
This does not mean the experimentation lacks value. It means people may be using AI for different purposes, applying different standards and learning lessons that remain largely invisible to the wider organization.
Possible impact: AI activity can expand without a shared understanding of the outcomes it is intended to improve. Local learning can remain disconnected, and leaders may have limited visibility into whether AI is improving the work or simply increasing its use.
Faster production is shifting the work of verification
Several people described producing analysis, written work or code much faster with AI. They also encountered uneven quality, questionable information and growing volumes of work requiring review by experienced employees.
Some initial tasks became faster. Verification, correction and accountability did not disappear. In some cases, the work moved.
Possible impact: One person may save time while another absorbs the verification. Experienced employees can become the quality-control layer for a growing volume of AI-enabled work, while the additional review, rework and risk remain largely invisible in productivity measures.
Improved output can conceal capability gaps
AI can help less experienced employees produce work that appears more complete before they have developed all the knowledge required to evaluate it.
Experienced colleagues may then carry the work of reviewing the output, correcting problems and supporting the person's development.
One person also questioned whether regular reliance on AI was weakening their own thinking. This was an individual concern, not a shared finding, but it points to an important tension between immediate performance and longer-term capability.
Possible impact: Organizations may confuse improved output with improved capability. If some of the experiences through which expertise was previously developed are removed, the organization may become more dependent on experienced people while weakening the pathway through which future experts are formed.
Individuals are absorbing the organizational work of AI experimentation
The people I spoke with are making their own decisions about how to use AI, what to trust, how to verify its output and how to learn.
They are also encountering existing constraints, including inaccessible data, technology bottlenecks, unclear authority, limited development capacity and uncertainty about what AI may mean for jobs. In several cases, deeper learning would need to happen outside already demanding working days.
These are not simply individual issues. They reflect the organizational environment in which people are experimenting.
Possible impact: Organizations may be treating AI experimentation as an individual responsibility while leaving the surrounding conditions largely unchanged. This can create uneven capability, hidden workload and anxiety. It can also cause new technology to reproduce existing limitations.
When people do not know what they can try, what they are accountable for or whether it is safe to disclose where AI failed, the organization may lose access to the learning it most needs.
What these signals may mean
AI is not only changing tasks. It is redistributing work, risk, accountability and learning.
The strongest signal across these conversations is that individuals are absorbing organizational work that has not yet been deliberately designed. They are interpreting the purpose of AI, deciding what to trust, verifying quality, managing risk, supporting colleagues and learning how to use it, often without shared expectations or protected capacity.
An organization can appear to be experimenting with AI while failing to learn from its own experimentation.
More consequential questions may need to be asked:
- What outcome is the work intended to improve, and for whom?
- Where is work being removed, and where is it being transferred?
- Who is carrying verification, risk and accountability?
- How are people developing the judgment required to evaluate AI-enabled work?
- What are people learning locally, and how will the organization see, compare and act on that learning?
Where this inquiry leads
Leaders do not need to redesign the entire organization at once. They do need to make the changing work visible.
One place to begin is a consequential workflow. The organization can examine how the work happens, how AI is changing it and where people, AI and organizational systems should contribute. A bounded test can then produce evidence about whether the redesigned workflow improves the intended outcome under real conditions.
One workflow cannot establish how the wider organization should change. It can surface constraints, dependencies and questions that warrant attention or further testing.
The redesign process also develops capability. Participants learn a repeatable way to examine work, test a different configuration and act on evidence. Tracking what is being learnt across workflows can help leaders distinguish what is local, what recurs and what may eventually justify broader change.
AI is already in the work.
The immediate leadership task is not simply to increase its use. It is to make the changing system visible before work, risk and learning settle into patterns no one deliberately chose.