Starting out01
How to get started when you need experience
Choose a realistic entry route, make one small work sample and ask for feedback. You do not need to master every AI tool first.
Issue 01 · September 2026Work beyond
Work beyond
the obvious.
A field guide to AI, robotics and work.
What the research says—and a practical
next step you can test.
Concept illustration · Human–machine collaboration
Our working argument · Editorial judgment
Better tools don’t settle the future of work.
AI can change a task without replacing a whole job. What happens next also depends on demand, costs, job design and who has the power to make decisions.
Our practical advice: understand the tools, learn a domain, and test whether your work improves an outcome. That can help you make better choices. It is not a guarantee of job security.
Your career journeyNot sure what to do next?
Not sure what to do next?
Start where you are.
Answer a few questions about your work, AI experience and available time. Get one practical next step and reading that fits. Come back with what happened, and your plan adjusts.
No account or coding required. A plan you can revise, not a promise of an AI-proof career.
For experienced workers after a layoffYour role ended.
Your role ended.
Your experience stays.
Work out what needs attention now, put your achievements into words and choose a next direction. Build a practical 30 / 60 / 90-day strategy around the experience you already have.
Find a way forward →Immediate priorities · Three routes to explore · A plan to take with you
Choose your way inDifferent moment.
Different moment.
Different first move.
Three first moves · Starting out
- Find a realistic entry-level role
- Make a small sample and ask for feedback
- Keep applying while you learn
Evidence, in context · Reviewed September 5, 2026Read the measure.
Read the measure.
Then the headline.
Potential task exposure, measured performance and projected job demand answer different questions. None gives your personal odds of losing a job.
Exposure is not displacement.
≈25%of global employment is in occupations with some generative-AI exposure.
- Highest exposure: 3.3% of all employment
- Other exposed categories
- Remainder outside these exposure categories
02 · Task performance · Different experiments
Benefits depend on the setting.
+26.08%
completed developer tasks
Three randomized field trials; 4,867 developers. Published February 2026, studying earlier coding assistants. Standard error: 10.3 percentage points.
Cui and colleagues · Study details →19% longer
to complete coding tasks
16 experienced open-source developers; 246 tasks with early-2025 tools. A small, specialized randomized study.
METR · July 2025 →Important 2026 update: METR’s follow-up moved toward speedups, but selection and measurement problems prevented a reliable current estimate. Read the follow-up →
These are not opposing estimates of one common rate. Tasks, tools, participants and outcome measures differ. We do not average them.
03 · United States · 2025–35 projectionFast growth.
Fast growth.
Or many new jobs?
A small occupation can grow quickly and still add relatively few jobs. Switch the measure to see why the distinction matters.
These roles also differ in pay, credentials, location and working conditions. Growth is not AI immunity or a recommendation to retrain.
Sorted by percentage growth. Nurse practitioners lead this selection.
View both measures as a table
| Occupation | Growth | Net new jobs |
|---|---|---|
| Nurse practitioners | 41.0% | 137,800 |
| Solar photovoltaic installers | 36.5% | 11,300 |
| Data scientists | 34.6% | 95,400 |
| Wind turbine service technicians | 29.5% | 3,500 |
| Home health & personal care aides | 18.1% | 847,300 |
Put the argument to work
Follow the handoffs.
A finished draft is not a finished service. Make the inputs, checks and decision owner visible, then find out whether the whole workflow improves.
One request. Four responsibilities. This is a design example, not a measured allocation of human and machine work.
- 01
Clarify
Person defines the problemFind the order and the customer’s actual request.
Verified records - 02
Draft
AI may assistPrepare a response from approved records and policy.
Draft + sources - 03
Verify
Person checks the evidenceCheck the facts, policy fit and any exception.
Checked recommendation - 04
Decide
Authorized person actsApprove the refund or escalate, then follow the result.
Outcome + corrections
Feed corrections back into the instructions, checks and policy. A named reviewer needs evidence, time and authority—not just an approval button.
Essential briefings
See the full desk Start where you are.
Changing course02
How to change direction without starting over
Use what you already know to explore a nearby role. Check the work, the pay and the route in before committing to retraining.
Robotics03
Is hands-on work a good next move?
Explore practical and robotics-related work through real employers and supervised training—not a list of supposedly robot-proof jobs.
Interactive field tool · Editorial guidancePlan one task.
Plan one task.
Not your whole future.
Think of a specific task. Answer three questions to sketch a trial, a human check and a stop rule. This is a planning aid, not a job-risk score.
Try an example
Your task: choose one activity you know. ‘Not sure’ is a valid answer.
Your working plan
Define the task before automating it.
At least one answer is still unknown. Agree on the task, the check and the consequences first.
- AI assistance
- Use a tool to brainstorm questions or outline a process with non-sensitive examples.
- Person checks
- Ask someone who knows the workflow to define a correct result and who can approve it.
- Stop or escalate
- Do not send outputs into live work until the unknowns have an owner and an answer.
Small trial
Walk through one completed case together. Write down the input, expected output, available check and failure consequence.
The rules are editorial, not empirically calibrated. Use approved tools and data. Resolve privacy, safety and authorization requirements before a live trial. A small comparison does not prove a causal effect.
How the guidance is chosen
Meaningful consequences take priority. Otherwise, unknown answers call for clarification; expert checks call for review capacity; variable tasks call for a narrower scope. Only clear, directly checkable tasks with contained consequences suggest a bounded pilot. No answer predicts employment or guarantees safe deployment.
Why task exposure is not a job-risk estimate →Reading feed · Automatically collected
On the reading list.
Links collected by our scheduled feed. Inclusion is not an endorsement or a completed evidence review. For the sources behind our claims, use the research library.
01MIT News · AI · Oct 2026Learner collaborates with MIT peers to develop book of AI insightsPrabha Jha used her experience with MIT Open Learning’s program for senior executives to develop AI solutions for her employer, Verizon, and to write a book.02MIT News · AI · Oct 2026Q&A: The “collision” between AI, your brain, and societyIn a new book, Professor Eric So explores how the technology affects our thinking skills.03MIT News · AI · Oct 2026MIT announces the MIT for America initiative, to strengthen STEM education across the countryThe effort aims to help U.S. learners from kindergarten to community college, with an emphasis on math, making, and the constructive use of AI.04U.S. Bureau of Labor Statistics · Aug 2026BLS adds occupation-level AI exposure categoriesA new four-category framework compares relative theoretical and observed AI exposure while warning that exposure is not an employment, wage, adoption, replacement, or productivity forecast.05World Economic Forum & PwC · Jun 2026A framework for reinventing entry-level workMore than one in three young workers globally are in occupations with medium-to-high exposure to AI-driven task change, raising questions about job access, design, and talent pipelines.
Feed checked November 3, 2026. Source descriptions may be lightly trimmed for length.