Teach people how to define an outcome, recognize an agent's limits, and decide when a person must approve the next action before expanding automation.
- 01Describe
State the outcome and permitted information.
- 02Delegate
Separate drafting from actions.
- 03Question
Recognize missing context and exceptions.
- 04Learn
Discuss the result and improve together.
Illustrative editorial framework, not measured research findings.
A new interface still needs informed people
The current wave of enterprise AI promises a simpler way to work: describe an outcome and let software coordinate the steps. That changes what employees need to learn. Navigating menus remains useful, but so does understanding what a request authorizes, which information an agent can use, and how to recognize an unfinished task. A conversational interface can reduce friction while leaving important judgments with the person using it.
This is practical analysis of developments announced earlier this month, rather than a claim that a new product launched today. On October 13, Workday described plans to bring Sana's agentic experience into tools including Claude, Gemini Enterprise, Microsoft Copilot, Teams and Slack. Its announcement distinguishes existing self-service capabilities, early access and functionality expected later in 2026. Organizations should check the availability of their specific capability rather than treating the entire announcement as immediately deployed.
Sources: [1]
What the adoption signals do and do not show
OpenAI's October 14 EU report describes growing use of agentic tools among its business customers. It also highlights uneven adoption and the importance of skills and confidence. These are observations from the company's own customer sample. They do not establish how every organization uses AI, or prove that more usage automatically produces better business outcomes. For leaders, the useful question is whether staff can turn access into responsible, useful work.
Independent reporting provides a complementary people perspective. IT Pro's October 16 discussion of workplace agents and copilots focuses on understanding how AI works and where it is weak. That discussion is context, not independent proof of any vendor's performance claims. Taken together, these sources support a timely management question: what should an employee understand before delegating work to an AI agent?
Teach delegation through one realistic task
The following approach is an editorial recommendation, not a prescribed vendor training program. Start a workshop with one familiar task and synthetic information. A hypothetical nonprofit might prepare a volunteer orientation pack; a small business might draft an internal project handover. Ask participants to state the audience, required information, desired result and deadline. Then ask them to identify what the agent should leave unresolved rather than guess.
Have learners separate drafting from action. Preparing a message, changing a record and sending a message are different permissions. Let participants practice requests that stop at a draft, then explain what would need to change before an external action became appropriate. This makes the boundary tangible without requiring employees to become developers or security specialists.
Include a deliberately incomplete example. Perhaps the handover notes omit an owner, or the orientation pack contains two conflicting dates. The learning objective is to spot missing context and ask for clarification. Reward the colleague who recognizes uncertainty as well as the colleague who writes an effective instruction. That encourages thoughtful use instead of a race to produce the most polished output.
Give managers a learning role
Training should also help managers design the surrounding work. Name someone who can answer questions, decide which information is permitted and handle exceptions. Set aside a short discussion after the first exercises: where did participants hesitate, which words were ambiguous, and which step needed human attention? Those observations can reveal a process problem that buying another tool would not solve.
For an individual learner, a useful practice is to explain the workflow back in plain language: what the agent received, what it attempted, what changed and what remains to be done. For organizations, use that explanation as a learning checkpoint before widening access. Success means a person can supervise the work confidently, not simply repeat a prompt from a presentation.
Begin with a modest session and a clear next step. Choose one routine task, create a safe example, and ask a small group to practice delegation and exception handling together. Keep outcomes visible and make it easy to raise concerns. AI transformation becomes more achievable when employees have permission to learn, ask questions and improve the process alongside the technology.
Sources and further reading
- Workday to Bring Sana Everywhere Employees Work ↗Workday · Accessed 2026-10-19
- The EU’s next chapter in AI is agentic ↗OpenAI · Accessed 2026-10-19
- Have agents, copilots, and chatbots made life harder? ↗IT Pro · Accessed 2026-10-19
AI-assisted research and writing for The Adaptable Mind. Analysis reflects an editorial interpretation of the linked sources, not a claim of personal testing or endorsement by the organizations mentioned.
Put the idea into practice.
Explore the free AI Basics series or discuss a learning or adoption challenge.
Start a conversation ↗