Prepare for changing work by examining the tasks you perform, the quality people need from them, and the skills that let you check the result. A small, measured experiment can tell you more about your next step than a prediction about an entire profession.
AI assistance, automation and new ways to learn can change how work is organized. Their value depends on the task, available data, people’s experience, and the cost of checking and maintaining the system. Start with your own workflow and a result you can recognize as good.
Read the evidence at the task level
The ILO’s 2025 occupational-exposure analysis estimates potential exposure to generative AI by examining tasks within occupations. Exposure describes what technology could affect; actual change also depends on cost, infrastructure, skills and how organizations implement it.
A study of AI assistance for customer-support agents found productivity gains that varied substantially with experience. That is evidence about a defined workplace and tool. It gives a reason to test similar assistance, while leaving other professions, tools and long-term outcomes open.
Expect the most useful questions to become more specific: Which parts of a report can be drafted reliably? Who checks exceptions? Does assistance improve a novice’s understanding? Does the organization use saved time to improve service, reduce workload, or change staffing? Those are separate outcomes.
Map a week of work
On a small screen, scroll the table sideways. Keyboard users can focus the table region and use the arrow keys.
| Task type | Possible assistance | Quality check that stays visible |
|---|---|---|
| Finding information | Search, retrieval or document comparison. | Open the original source and verify coverage, dates and quotations. |
| Drafting a routine output | Create a first version from an approved template. | Check facts, audience, completeness and handling of exceptions. |
| Calculating or coding | Propose formulas, tests or a small script. | Use known inputs and independently expected results. |
| Making a consequential decision | Organize options and evidence. | A responsible person assesses context, uncertainty and consequences. |
| Teaching or learning | Offer hints, practice questions and feedback. | Solve a new problem independently and explain the reasoning. |
For each task, record frequency, usual duration, input sensitivity and who uses the output. Select something reversible, bounded and easy to inspect. Use approved tools and data. Where the work involves personal records, confidential material or regulated decisions, establish the organization’s rules before experimenting.
Measure useful work, including the checking
- Keep several ordinary examples as a baseline, including a difficult one.
- Define acceptance criteria before testing: accuracy, completeness, useful format and total time.
- Use the same type of input and record the tool/version and assistance provided.
- Have someone inspect the result against the original sources.
- Try another set of examples, including exceptions. Keep, revise or stop the workflow based on the evidence.
A pilot also needs an exit path. Save usable files in a format the team can maintain, document the checking step, and identify who will review changes when the tool updates.
Make assistance produce learning
A 2025 randomized study in an undergraduate physics course found stronger immediate learning outcomes with a deliberately designed AI tutor than with the comparison classroom lessons. The setting involved specific material, instructional design and assessments. It supports investigating structured tutoring, with retention and transfer to new problems still needing attention in your own learning.
Use assistance to ask for a hint, explain a mistake, or generate a practice variation. Then close the assistance and solve another problem. Returning later to explain the idea from memory gives you a different kind of evidence from recognizing a polished answer on screen.
- Before: attempt a short task independently and record what you can already do.
- During: ask for feedback on your reasoning and compare it with a trusted reference.
- After: complete a fresh task without hints and explain why the method works.
- Later: repeat with a changed context to check retention and transfer.
Choose skills and signals worth following
Build a combination of subject knowledge, source checking, communication and the ability to design a small test. These skills help you evaluate new tools as the tools change. Practical experience with the people and systems affected by your work makes abstract output easier to judge.
Watch for changes in real task requirements, employer training, supported workflows and quality measures. A tool’s new benchmark score is one signal; a team maintaining good results over repeated use is a stronger adoption signal. Look for who benefits, who takes on extra checking, and whether access to training is shared.
Choose one two-week learning experiment. Write the task, evidence of success, time available, data boundaries and a stop condition. At the end, decide whether to continue, adapt or try something else. That gives you a concrete preparation for several possible futures.
Use the worksheet
Download the task and skills worksheet. Save the Markdown file and edit it in a text editor or Obsidian. It includes instructions and space for your own notes.
Continue exploring
- Judge signals and predictions
- Test a learning decision across scenarios
- Explore education resources
- Explore the Future library
Evidence and links reviewed October 17, 2026. Future developments remain conditional on the evidence and circumstances described.