What can AI actually do for my job?

Job titles differ, but the tasks underneath overlap heavily. Below is a list organized by how much supervision each type of task needs, as of September 2026.

What research says about exposure

• A study posted on arXiv estimated that around 80% of the U.S. workforce could have at least 10% of their work tasks affected by LLMs, while approximately 19% of workers may see at least 50% of their tasks impacted. The arXiv paper by Eloundou et al. says its authors do not make predictions about the development or adoption timeline of LLMs.
• An NBER working paper that classified ChatGPT conversations found that “Practical Guidance,” “Seeking Information,” and “Writing” are the three most common topics, together nearly 80% of conversations. That matches the first list below.
• Anthropic’s Economic Index reports that 49% of jobs in its sample saw Claude used for at least a quarter of their tasks, up from 36% in January 2025.

What built-in assistants list, by app

• Microsoft Learn lists Word features in Microsoft Copilot: draft, rewrite, and summarize documents.
• Microsoft Learn lists Excel features in Microsoft Copilot: analyze data, generate insights, create formulas and visuals.
• Microsoft Learn lists PowerPoint features in Microsoft Copilot: create presentations from prompts or existing content, and summarize or edit presentations.
• Microsoft Learn lists Teams features in Microsoft Copilot: summarize meetings for up to 30 days, transcribe them, and capture action items.
• Google lists Gemini in Gmail jobs such as summarizing long email threads into concise bullet points and drafting or refining emails for tone and clarity.

Tasks AI handles well

Writing and editing. First drafts of emails, memos, proposals and status updates; rewriting for tone or length; turning bullet points into prose or the reverse. See writing emails with AI.

Summarizing. Long reports, contracts, email threads and transcripts reduced to key points, decisions and open questions. See summarizing long documents.

Reformatting. Turning messy notes into a table, converting a document to a different template, extracting names and dates into a list.

Explaining. Asking “what does this formula do?” or “explain this clause in plain English” saves a lot of searching.

Tasks AI does with supervision

Spreadsheet work. Cleaning data, writing formulas, spotting anomalies and building summaries. Always reconcile a total or a row count against the original. The Excel guide shows how.

Research. Gathering background on a market, competitor or topic. Ask for sources and open them; assistants can produce plausible but wrong citations.

Meeting notes and action items. Good when there is a transcript; you still confirm owners and due dates.

Contract review. Flagging unusual clauses and comparing against your standard terms is useful as a first pass, not a replacement for a lawyer where it matters. See whether AI can review contracts.

Automation. Agent-style tools such as Claude Code can perform multi-step work on your files, like producing a weekly report from a folder of exports. They need clear rules and a review step.

Tasks to keep for yourself

• Final decisions on hiring, firing, pricing, legal positions or medical or financial advice.
• Relationship-heavy communication where your voice and judgment are the point.
• Anything involving data you are not permitted to share with an AI service.

How to find what applies to you

List what you did last week hour by hour. Mark tasks that were reading, writing, formatting or looking things up. Those are the candidates. Then estimate how long each takes and how easy it is to check the result. High time, easy to check: start there.

An important caveat

AI output sounds confident whether it is right or wrong. The more a task depends on facts the assistant does not have, the more it will fill gaps with plausible guesses. Provide the source material yourself instead of relying on memory, and read how to reduce hallucinations.

Learn AI Fast teaches the general method: on day two, everyone maps recurring tasks for the same case company using a standard worksheet, then applies the method to their own job afterward.

Sources

• arXiv, GPTs are GPTs (Eloundou et al.): “around 80% of the U.S. workforce could have at least 10% of their work tasks affected”
• arXiv, GPTs are GPTs (Eloundou et al.): “We do not make predictions about the development or adoption timeline”
• NBER, How People Use ChatGPT: “are the three most common topics and collectively account for nearly 80% of all conversations”
• Anthropic, Economic Index: New building blocks for AI use: “36% of jobs in our sample saw Claude being used for at least a quarter of their tasks. Pooling data across reports, this has risen to 49%”
• Microsoft Learn, What is Microsoft Copilot?: “Draft, rewrite, and summarize documents”
• Microsoft Learn, What is Microsoft Copilot?: “Analyze data, generate insights, create formulas and visuals”
• Microsoft Learn, What is Microsoft Copilot?: “Create presentations from prompts or existing content, summarize presentations or edit presentations”
• Microsoft Learn, What is Microsoft Copilot?: “Summarize (up to 30 days) and transcribe meetings, and capture action items”
• Google, Collaborate with Gemini in Gmail: “Synthesize long email threads into concise bullet points”