Plain-language summary
What this guide covers
AI may help workers draft, summarize, classify, search, organize, and plan. That does not mean a whole job is automated. These guides show how to look at tasks, workplace rules, data sensitivity, review needs, and human responsibility before using AI.
Work is made of tasks, relationships, records, tools, policies, and decisions. A tool that helps with one task can still create privacy, accuracy, fairness, access, or accountability problems if it is used in the wrong place. The role guides give readers a calm way to test low-risk uses while protecting customers, coworkers, students, patients, projects, and organizations.
What you will learn
- Distinguish occupational exposure from real workplace adoption and employment outcomes.
- Separate assistance, partial automation, and full automation in plain language.
- Use a task-level framework to decide where AI is low risk, high risk, or unsuitable.
- Identify privacy, security, accessibility, and decision-ownership questions before using AI at work.
- Choose the most relevant role guide and connect it to skill-building routes.
Guide section
How to use these role guides
These guides are built for nontechnical readers who want to understand where AI may help at work without turning a job into a prediction contest.
Start with the role closest to your daily work, then read the task map before trying any prompt. The main idea is simple: AI changes parts of workflows before it changes job titles. A tool may draft, summarize, classify, compare, or format information, but the worker and organization still own the result. These pages separate occupational exposure from actual adoption. Exposure means a task has features that an AI system might assist with. Adoption means a real workplace has approved, trained, monitored, and integrated a tool for that task. Those are different things. A polished AI output is not the same as a verified result. Individual experimentation is not the same as organization-approved deployment.
A safe reading path
- Read the plain-language summary and learning objectives.
- Find tasks you actually perform, not tasks that only appear in a formal job description.
- Check the risk level in the task map before experimenting.
- Compare the starting tasks with the unsuitable uses.
- Use the first-week experiment only with approved tools and low-risk information.
- Ask your employer the policy questions before using private, customer, personnel, legal, financial, health, contract, or security-related information.
- Review the related skills routes when a task needs better prompting, data literacy, communication, or judgment.
Guide section
A task-level framework for AI at work
The safest way to think about AI is to look at one workflow at a time and ask what the tool is doing, what the person is checking, and who owns the final decision.
Five levels of AI involvement
| Level | What AI does | Human responsibility | Use with care |
|---|---|---|---|
| Reference aid | Searches, retrieves, organizes, or restates information. | Confirm the source, context, date, and completeness. | Knowledge bases, policy summaries, meeting notes, and document outlines. |
| Drafting assistant | Creates a first draft of text, lists, tables, or messages. | Edit for accuracy, tone, audience, accessibility, and policy. | Emails, agendas, summaries, scripts, and status updates. |
| Partial automation | Completes a bounded step inside a workflow, often using rules or integrations. | Monitor inputs, outputs, logs, exceptions, and handoffs. | Ticket routing, calendar options, report formatting, and duplicate detection. |
| Decision support | Suggests options, risks, priorities, or next actions. | A responsible person decides and documents the reason. | Escalation triage, risk registers, account planning, and project tradeoffs. |
| Full automation | Acts without normal human review for a defined task. | Use only when approved, tested, monitored, reversible, and allowed by policy. | Routine notifications, standard reminders, and low-risk formatting jobs. |
Try it
Try a 15-minute task review
Pick one recurring task before choosing a tool. Write the task in one sentence, then list the inputs, people affected, records created, and final decision owner. Mark each input as public, internal, confidential, customer data, employee data, or regulated information. Then decide whether AI should be used at all, and if so, whether it should only draft, summarize, sort, or suggest. This keeps the focus on work design, not novelty. It also reveals hidden parts of a workflow: approvals, retention rules, accessibility needs, handoffs, and the moment when a person must verify the result.
- Name the task and the business purpose.
- List the information used and the records produced.
- Classify the data by sensitivity.
- Choose the allowed AI role: none, assist, partially automate, or fully automate under approved controls.
- Name the reviewer and final decision owner.
- Write one stop condition that would end the experiment.
Guide section
Safety and privacy baselines
AI safety at work is mostly ordinary discipline: use approved tools, protect data, keep humans accountable, and do not let fluent text pass as proof.
Baseline practices
- Use organization-approved AI tools. Public tools may not be allowed for internal, customer, personnel, financial, legal, health, contract, or security information.
- Minimize data. Remove names, account numbers, addresses, calendar details, confidential strategy, and other sensitive details when they are not needed.
- Verify against trusted records. AI can produce polished text that is wrong, outdated, incomplete, or invented.
- Label uncertainty. A useful draft should show open questions, missing information, and assumptions.
- Keep records according to policy. Some messages, decisions, tickets, meeting notes, or project records may have retention requirements.
- Make content accessible. Check headings, plain language, links, tables, reading order, captions, and alternatives for images or attachments.
- Do not use AI to make employment, legal, financial, medical, cybersecurity, or other high-impact decisions unless a qualified person and approved process own the decision.
Guide section
Choose the guide that fits your work
Job titles can hide very different workflows. Choose the guide based on what you do each week.
Role navigation
| Route | Best fit | Start with this question |
|---|---|---|
| /roles/administrative-assistant | You coordinate calendars, documents, records, inboxes, meetings, travel, forms, or office follow-up. | Which recurring administrative task could be drafted, organized, or checked with low-risk data? |
| /roles/customer-success | You help customers after purchase, respond to tickets, summarize conversations, prepare account notes, or spot feedback themes. | Where could AI help you find information faster without making unauthorized promises? |
| /roles/project-manager | You coordinate plans, meetings, risks, dependencies, status reports, decisions, and lessons learned. | Where could AI reduce documentation friction while people keep ownership of scope, priority, budget, and performance decisions? |
| /roles/teacher | You plan lessons, adapt materials, explain topics, give feedback, or support accessibility. | How can AI support preparation while preserving educator judgment and student privacy? |
| /roles/data-analyst | You clean, interpret, chart, or explain data. | How can AI help document assumptions while you verify calculations and data quality? |
| /roles/software-engineer | You write, review, test, document, or debug code. | How can AI help with low-risk drafts while you protect secrets and run tests? |
| /roles/marketing-coordinator | You draft campaigns, calendars, briefs, social posts, or research summaries. | How can AI speed drafts while a person checks claims, brand, rights, and audience impact? |
| /roles/sales-representative | You prepare outreach, research accounts, write follow-ups, and update CRM records. | How can AI help with preparation without deceptive personalization or unauthorized commitments? |
| /roles/healthcare-administration | You support scheduling, records, billing workflows, front-desk service, or referrals. | How can AI help routine admin while protecting health information and clinical boundaries? |
| /roles/small-business-owner | You juggle operations, customers, finance, marketing, vendors, and planning. | Which narrow workflow needs a checklist before automation? |
Guide section
What research can and cannot tell us
AI workforce evidence is useful, but it is not one kind of evidence. The safest reading is to compare studies by what they measured.
Evidence types
| Evidence type | What it can show | What it cannot prove by itself |
|---|---|---|
| Task exposure analysis | Which tasks have features that current models may assist with. | Whether a workplace will adopt a tool or whether employment will rise or fall. |
| Employer surveys | What surveyed employers expect, plan, or worry about at a point in time. | What actually happened across the whole labor market. |
| Worker surveys | How workers report tool use, trust, training, and concern. | Whether reported use is safe, approved, or effective. |
| Field studies | What happened in one real setting with one tool and workflow. | That the same result will occur in all roles, teams, countries, or tools. |
| Official labor projections | Employment projections by occupation and geography under stated assumptions. | The effect of AI alone or the future of one individual worker. |
As of June 20, 2026, several findings are useful but limited. The ILO’s 2023 global analysis framed generative AI as more likely to augment many tasks than fully automate whole occupations, while noting higher exposure in clerical work and important differences by country and gender. The OECD reports that workplace outcomes are shaped by training, worker consultation, and implementation choices. The 2025 World Economic Forum report is an employer survey across many economies, useful for expectations but not the same as observed outcomes. A 2023 customer-support field study found productivity gains in one setting with one tool and workflow, especially for newer workers, but that result should not be treated as a promise for every team.
Avoidable errors
Common mistakes and better approaches
Asking whether a whole job is safe from AI.
Better approach: Ask which tasks can be assisted, which tasks require review, and which tasks are unsuitable for AI.
Treating a confident AI answer as verified.
Better approach: Check the answer against approved records, current policy, and a responsible reviewer.
Using public tools with private workplace data.
Better approach: Use only approved tools and minimize sensitive data.
Skipping accessibility and tone review.
Better approach: Review reading level, structure, alternatives, and audience impact before publishing or sending.
Assuming a study or survey applies everywhere.
Better approach: Check geography, industry, date, method, and what the study actually measured.
Remember this
Key takeaways
- AI affects tasks before it affects job titles.
- Exposure is not the same as adoption or employment outcome.
- Assistance, partial automation, and full automation have different review needs.
- A polished AI output still needs verification.
- Approved deployment is different from individual experimentation.
- Human decision ownership should be clear before a workflow changes.
- Privacy, security, accessibility, and fairness belong in the first conversation, not the last.
Questions readers ask
Frequently asked questions
Does occupational exposure mean my job will disappear?
No. Exposure means some tasks have features that AI may assist with. It does not prove that a tool will be adopted, that it will work reliably in your workplace, or that employment outcomes will change in a specific way.
What is the difference between assistance and automation?
Assistance means AI helps a person draft, organize, summarize, or compare information. Partial automation means AI completes a bounded step inside a workflow, often with monitoring. Full automation means the system acts without normal human review for a defined task and should require stronger approval, testing, monitoring, and rollback.
Can I paste work documents into any AI tool?
Not safely. Use only approved tools and follow your organization’s rules for confidential, customer, personnel, legal, financial, health, security, and contract information.
What skill matters most?
Verification. Prompting helps, but the safer worker knows how to check facts, sources, dates, tone, policy, records, and decision ownership.
Should every team use AI the same way?
No. Teams differ by data sensitivity, customer impact, regulation, accessibility needs, contracts, records, culture, and approved tools. The same task can be low risk in one setting and high risk in another.
Sources and review notes
Sources were accessed on the dates shown. Links open the original organization’s page.
- SRC-01Secretaries and Administrative Assistants, Except Legal, Medical, and Executive (43-6014.00)U.S. Department of Labor, O*NET OnLine · Accessed 2026-06-20
- SRC-02Customer Service Representatives (43-4051.00)U.S. Department of Labor, O*NET OnLine · Accessed 2026-06-20
- SRC-03Project Management Specialists (13-1082.00)U.S. Department of Labor, O*NET OnLine · Accessed 2026-06-20
- SRC-04Secretaries and Administrative Assistants: Occupational Outlook HandbookU.S. Bureau of Labor Statistics · Published 2025-08-28 · Accessed 2026-06-20
- SRC-05Customer Service Representatives: Occupational Outlook HandbookU.S. Bureau of Labor Statistics · Published 2025-08-28 · Accessed 2026-06-20
- SRC-06Project Management Specialists: Occupational Outlook HandbookU.S. Bureau of Labor Statistics · Published 2025-08-28 · Accessed 2026-06-20
- SRC-07Generative AI and Jobs: A global analysis of potential effects on job quantity and qualityInternational Labour Organization · Published 2023-08-21 · Accessed 2026-06-20
- SRC-08Artificial Intelligence Risk Management Framework (AI RMF 1.0)National Institute of Standards and Technology · Published 2023-01-26 · Accessed 2026-06-20
- SRC-09Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence ProfileNational Institute of Standards and Technology · Published 2024-07-26 · Accessed 2026-06-20
- SRC-10AI and workOrganisation for Economic Co-operation and Development · Accessed 2026-06-20
- SRC-11The Future of Jobs Report 2025World Economic Forum · Published 2025-01-07 · Accessed 2026-06-20
- SRC-12AI Companies: Uphold Your Privacy and Confidentiality CommitmentsFederal Trade Commission · Published 2024-01-09 · Accessed 2026-06-20
- SRC-13Web Content Accessibility Guidelines (WCAG) 2.2World Wide Web Consortium · Published 2023-10-05 · Accessed 2026-06-20
- SRC-14Department of Labor releases AI Best Practices roadmap for developers, employers, building on AI principles for worker well-beingU.S. Department of Labor · Published 2024-10-16 · Accessed 2026-06-20
- SRC-15EEOC Launches Initiative on Artificial Intelligence and Algorithmic FairnessU.S. Equal Employment Opportunity Commission · Published 2021-10-28 · Accessed 2026-06-20
- SRC-16Generative AI at WorkNational Bureau of Economic Research · Published 2023-04-01 · Accessed 2026-06-20
- SRC-17GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language ModelsarXiv; authors affiliated with OpenAI, OpenResearch, and University of Pennsylvania · Published 2023-08-21 · Accessed 2026-06-20