The Shift: A Baseline Skill

In 2023, using AI at work made you an early adopter. In 2026, not using it requires explanation. AI literacy — knowing which tasks AI handles well, how to instruct it clearly, and when to stay skeptical — is now a professional skill comparable to knowing how to use a spreadsheet or write a coherent email.

This does not mean AI does your job. It means AI handles certain tasks within your job faster than you could by hand. The skill is knowing which tasks those are, and how to direct AI to do them well.

Five categories account for the vast majority of productive AI use in professional settings. Each has a distinct set of working prompts and real limitations.

1. Writing: Drafting, Editing, Summarizing

Writing is the highest-adoption AI use case at work because the feedback loop is fast and the output is immediately evaluable. You can read the draft in 30 seconds and know whether it's useful. The key shift: AI writes the first draft, you edit and own the final version.

✍
Writing at Work
Drafts, edits, summaries, reformats

Use AI to generate a working draft, then edit for accuracy, tone, and any information the model couldn't know. AI drafts tend to be structurally sound but generic — your job is to add the specific facts, voice, and judgment the model lacks.

Prompt: First draft from bullet points
I need to write a [type of document: status update / proposal / summary] about [topic]. Here are my raw notes: [paste bullet points] Write a first draft in clear, professional language. Aim for [length: 200 / 400 / 800 words]. Avoid filler phrases and hedging language. I will edit for accuracy after.
Why this works: giving raw notes lets the model structure your thinking without inventing facts.
Prompt: Editing for clarity
Edit the following text for clarity and concision. Remove redundant phrases, shorten sentences over 25 words, and keep the meaning exactly as written. Do not change the substance or add new information. [paste your draft]
Why this works: constraining the model ("do not add") prevents it from inventing content during the edit.
Prompt: Summarizing a long document
Summarize the following document in [3 bullets / 1 paragraph / 5 key points]. Focus on: [what decision needs to be made / what the reader must act on / the core argument]. Do not include background context the reader doesn't need. [paste document]
Why this works: specifying the focus prevents the model from summarizing the preamble instead of the substance.

2. Research: Synthesizing, Fact-Checking, Competitive Intel

AI accelerates research by synthesizing background knowledge quickly. It is not a replacement for primary sources on current or sensitive topics — AI models have knowledge cutoffs and can confabulate. Use AI for orientation and hypothesis generation, then verify with primary sources.

🔎
Research at Work
Background synthesis, comparison, structured exploration

AI is most reliable for synthesizing established knowledge — how a process works, how competitors are positioned, what the key tradeoffs in a decision space are. It is less reliable for recent events, specific statistics, and niche domain claims. Treat AI research output as a starting framework, not a finished product.

Prompt: Structured background on an unfamiliar topic
I need to get up to speed on [topic] quickly. Give me: 1. A 2-sentence plain-language explanation of what this is 2. The 3-5 most important concepts I need to understand 3. The main tradeoffs or debates in this area 4. What I should read or search to go deeper I have [X] background in this area. Calibrate the explanation accordingly.
Why this works: structured output prevents a generic overview dump; calibration prevents over-explaining basics you know.
Prompt: Competitive landscape summary
Summarize the competitive landscape for [product/market category] as of your knowledge cutoff. For each major competitor, cover: positioning, key strengths, known weaknesses, and the customer segment they serve best. Flag anything you're uncertain about or that may have changed recently.
Why this works: asking the model to flag uncertainty gives you a signal for where to do primary research.

3. Data Analysis: Interpreting, Querying, Explaining

AI is useful for data analysis in two ways: interpreting what numbers mean in plain language, and writing the code (SQL, Python, Excel formulas) to extract or transform data. AI does not run queries against your live database unless integrated with tools that enable that — it writes the code, you run it.

📊
Data Analysis at Work
Interpretation, code generation, explanation

Non-technical professionals can get the most value from AI data work by using it to interpret results and write code they can paste and run. Technical professionals use it to accelerate writing boilerplate SQL and data transformation code. In both cases: verify the output before acting on it.

Prompt: Interpret a table or chart in plain language
Here is a data table from our [report / dashboard / export]: [paste table or describe the data] Explain what this data shows in plain language. What is the most important thing to notice? Are there any patterns, anomalies, or potential concerns? What follow-up questions would a thoughtful analyst ask?
Why this works: asking for follow-up questions surfaces what you may not have thought to look for.
Prompt: Write a SQL query
Write a SQL query to [describe what you want to find]. My table is called [table_name] and has these columns: [list columns with types if known]. I want the output sorted by [column] and limited to [N] rows. Use standard SQL — this runs on [PostgreSQL / BigQuery / MySQL].
Why this works: specifying the SQL dialect prevents syntax incompatibilities. Always test on a sample before running on full data.
Prompt: Explain a metric or result
Our [metric: churn rate / conversion rate / NPS / CAC] is [value]. Help me understand: (1) whether this is typical for a [type of business], (2) what factors most commonly drive it up or down, and (3) what are the 3 most likely causes if it's outside the normal range. Flag any assumptions you're making.
Why this works: asking for flagged assumptions makes the model more careful and gives you a checklist to verify.

4. Communication: Emails, Meeting Prep, Follow-Ups

Communication tasks — especially ones that require diplomacy, follow-up timing, or adapting to a specific relationship — are where AI can save significant time. The model handles the structure and language; you supply the context and judgment about what this particular person or situation needs.

✉
Communication at Work
Emails, agendas, follow-ups, briefs

AI is particularly useful for communication tasks where you know what you need to say but the drafting takes disproportionate time — or where you need to find the right tone for a difficult message. Always read the draft as if you're the recipient before sending.

Prompt: Draft a difficult email
Draft an email from me to [recipient role, e.g. "a client who missed a deadline"]. The situation: [2-3 sentences of context]. The goal of this email: [what I want to happen as a result]. Tone: [professional / firm / empathetic / direct]. Length: [short / medium / keep it under 200 words]. I will edit before sending.
Why this works: specifying the goal (not just the situation) helps the model write toward an outcome, not just describe the problem.
Prompt: Prepare for a meeting
I have a [type of meeting: client check-in / stakeholder review / team retrospective] in [X] minutes on [topic]. Help me prepare: 1. The 3 most important questions I should ask 2. Potential objections or concerns the other party might raise and how to address them 3. A one-sentence framing statement to open the meeting Context: [brief background on the situation]
Why this works: asking for objections forces the model to think from the other party's perspective, not just yours.

5. Learning: Skill Development, Concept Explanation, Onboarding

AI is an effective on-demand tutor. It can explain concepts at the right level, generate practice questions, and create training materials — faster than building a course and more personalized than a generic tutorial. This is underused in most organizations.

🎓
Learning at Work
Explanations, practice, onboarding, skill development

The most effective learning use of AI: ask it to explain the same concept multiple ways until one clicks. This is something a human tutor would do but a book cannot. You can also use AI to generate practice questions, review your understanding, and create onboarding materials for new team members.

Prompt: Explain a concept you don't understand
Explain [concept] to me as if I'm a [your role: product manager / marketer / finance analyst] with no technical background in this area. Use an analogy if it helps. Then: (1) tell me the one thing most people get wrong about this, and (2) give me a simple way to test whether I've actually understood it.
Why this works: the "test your understanding" step turns passive reading into active recall.
Prompt: Create an onboarding document
I need to onboard a new [role] to our [process / system / product area]. Help me create a structured onboarding document covering: what they need to know in the first [week / month], the 5 most important things to understand about how we do this, common mistakes new people make, and who they should ask when they're stuck. Here's what I know about our process: [paste notes]
Why this works: AI structures your institutional knowledge into a usable document rather than letting it stay in your head.

What NOT to Use AI For

Appropriate skepticism is part of AI literacy. These are the categories where overreliance on AI output leads to real problems.

Avoid Real-time data

AI models have training cutoffs. Do not use a general-purpose AI to answer questions about current prices, recent news, today's regulatory status, or anything requiring data from the last few months. Use Perplexity or direct web search for current information.

Avoid Legal or medical advice without review

AI can explain legal and medical concepts helpfully, but should not substitute for licensed professional judgment on decisions with real consequences. Use AI to understand the landscape; use a professional to advise on your specific situation.

Avoid Sensitive HR decisions

Performance reviews, disciplinary documentation, hiring decisions, and compensation analysis involving specific individuals carry legal and ethical weight that AI cannot reliably navigate. AI can help you draft neutral language, but a human must review and own the output.

Avoid High-stakes outputs without verification

AI confabulates — it produces confident-sounding incorrect information. Never publish, file, or act on AI output that contains specific facts (statistics, citations, dates, names) without independently verifying those facts. The draft is a starting point, not a finished product.

The Skill Is Prompting

Across all five categories, one factor determines whether AI is genuinely useful at your job: how well you instruct it. This is the prompting gap — the consistent finding that the same tool produces dramatically different results depending on who is using it and how they're writing their prompts.

Three things account for most of the gap:

01
Specificity
Vague prompts produce generic outputs. The more precisely you describe the task, the audience, the length, and the format, the more useful the first draft will be. "Write an email" produces filler. "Write a 150-word follow-up email to a client who hasn't responded in 10 days, tone firm but not aggressive" produces something editable.
02
Context
AI cannot know your company, your industry, your relationship with the recipient, or your internal jargon unless you tell it. Providing 2-3 sentences of relevant context — even when it feels obvious — dramatically improves output quality. Context you leave out is context the model has to guess at.
03
Iteration
The first output is rarely the final output. Use follow-up instructions to refine: "Make this shorter," "Change the tone to more direct," "Remove the third paragraph," "Add a specific example." AI responds well to incremental refinement — don't start over, iterate from the draft.

The practical starting point: Pick one recurring task you do at least weekly that involves writing or summarizing. Use AI for that one task for two weeks. Measure the time difference. Then, once the habit is established in one place, expand to a second task. Trying to integrate AI into everything at once usually results in it being used for nothing consistently.

Stay current on AI tools,
workflows, and what actually works

The AI Rundown covers practical AI use at work — specific prompts, new tools, and honest assessments — free, every weekday morning.

Frequently Asked Questions

How do I start using AI at work?
Start with a task you do repeatedly that involves reading or writing text. Common starting points: summarizing long documents or meeting notes, drafting a first version of a routine email or report, researching an unfamiliar topic. Pick one task, use AI for 2 weeks, then evaluate the time saved before expanding. Trying to change all your workflows at once produces consistent results in none of them.
What work tasks are AI best suited for?
AI delivers consistent value in five categories: writing (drafting, editing, summarizing, reformatting), research (synthesizing background knowledge, comparing options), data analysis (interpreting numbers, writing queries), communication (drafting emails, meeting prep, follow-ups), and learning (explaining concepts, generating practice questions, creating onboarding materials). These categories share a common characteristic: the task involves text transformation where the right output is recognizable once you see it.
What should I NOT use AI for at work?
Avoid relying on AI for: real-time or current data (use search for current facts), final legal or medical decisions without expert review, performance evaluations of specific colleagues, and any high-stakes output with specific factual claims you haven't independently verified. AI confabulates — it can produce confident, detailed, incorrect information. Always verify specific facts before acting on them.
How do I write better AI prompts at work?
Three things improve prompts most: specificity (describe the task, audience, length, and format explicitly), context (give the model 2-3 sentences of relevant background it couldn't otherwise know), and iteration (use follow-up instructions to refine the draft rather than starting over). "Write an email" is a weak prompt. "Write a 150-word follow-up email to a client who missed a deadline, tone firm but professional" is a strong one.
Is it safe to share work documents with AI?
Check your company's AI usage policy before sharing any work documents. Enterprise AI tools (Microsoft Copilot, Claude for Enterprise, Google Workspace AI) typically have data handling agreements that prevent training on your content. Consumer products may have different policies. When uncertain: remove identifying information (client names, account numbers, financial specifics) before pasting into a consumer AI tool, then reinsert after editing the draft.
Will AI replace my job?
For most knowledge workers in 2026, AI is changing how work gets done, not eliminating roles. Tasks being automated are typically repetitive text processing — formatting, first drafts, summarization — rather than judgment, relationships, or deep domain expertise. The more realistic near-term effect: AI-skilled professionals can produce more output per hour, which raises expectations over time. Learning to use AI well is a professional skill with genuine value, regardless of the longer-term trajectory.