Your First Delegated Task
This content was developed with AI assistance and is regularly reviewed for accuracy.
Snapshot: October 2026. Model names, prices, and product features change monthly; What's New in AI is the maintained chronology.
This playbook walks you through handing an AI assistant a real task and getting back something you can use. Instead of typing a question and hoping, you will write a short brief, let the assistant search and read on your behalf, check what comes back, and keep the brief for next time. By the end you will have one finished task and a reusable way of asking.
Last tested with Claude (free plan) and ChatGPT (free plan) in October 2026.
What you'll accomplish
- Pick a task worth delegating and describe what "done" looks like
- Write a six-field brief instead of a one-line question
- Supply context by attaching a file and letting the assistant search
- Check the result and redirect it once, or restart with a better brief
- Save the brief so the next similar task takes two minutes
Prerequisites
- A computer or phone with internet access
- A free account on Claude, ChatGPT, or Gemini. Any of them works for this playbook. Free tiers give you the smaller model in each family rather than the flagship, and ChatGPT's free tier has shown ads since February 2026 (reported); the Free AI Tools Directory tracks what each plan includes.
- One real task you need done this week
- 20 minutes of focused time
No technical background required. If you can write an email to a colleague asking for help, you can do this.
The playbook
Step 1: Choose a task worth delegating
Goal: Pick something with a clear finished state that would otherwise take you 30 minutes or more.
Action: Good first tasks have an output you can hold up and judge: a comparison table, a plan with constraints, a document built from your notes, a summary of a long PDF. Skip two kinds of task for now. Decisions only you can make (whether to take the job) are yours to own; the assistant can lay out the options. And anything irreversible, such as sending a message to someone else or spending money, stays with you until you have checked a lot of its work.
A quick test: can you say in one sentence what "done" looks like? "A two-day plan with venue names and costs" passes. "Help me with my trip" does not.
Checkpoint: You have one task and a one-sentence definition of done.
Step 2: Write the brief
Goal: Give the assistant the same information you would give a capable new colleague.
Action: A brief has six short fields. You do not need headings in the message, but writing them out the first few times stops you skipping the two that beginners drop most often: resources and verification.
| Field | What to write |
|---|---|
| Objective | What you want done (not suggested, done) |
| Context | Who this is for, why it matters, what is already decided |
| Resources | The files, links, or searches it should use |
| Constraints | Budget, length, deadline, things to avoid |
| Output | The shape of the result |
| Verification | How it should show you the work is right |
Example:
Objective: Plan a two-day weekend in Austin, Texas for two adults who like live music and being outdoors.
Context: We arrive Friday evening and leave Sunday at 4pm. We're staying at the hotel in the attached confirmation. Budget is about $300 for activities and meals, not counting the hotel.
Resources: Search the web for current venue hours and this weekend's live music listings. Use the attached confirmation for our location.
Constraints: No more than three activities per day. Nothing that needs a car. One of us is vegetarian.
Output: A day-by-day plan with morning, afternoon, and evening blocks, venue names, rough costs, and walking or transit time between stops.
Verification: For each venue, give me the page you checked for its hours. Flag anything you could not confirm.
Two habits make briefs work in 2026. First, ask for the thing you want done. Current assistants follow instructions literally; "can you suggest some venues" gets suggestions, while "plan the weekend" gets a plan. Second, leave out the decoration. You do not need to assign an expert persona or ask it to think step by step: a December 2025 Wharton study found no consistent accuracy gain from expert personas, and current models reason before answering on their own. A role is fine for setting tone ("write this for a nervous first-time traveler"), not for correctness.
Before you send, run the colleague test from Anthropic's guidance: if a coworker with no context read this message, would they know what to do? If not, the assistant will not either.
Checkpoint: Your brief has all six fields and passes the colleague test.
Step 3: Supply the context and send
Goal: Let the assistant read and search instead of retyping what it needs.
Action: Attach the file your brief mentions with the paperclip or plus button; drag-and-drop works in all three apps. The assistant will read it directly. You do not need to paste the text in, and you do not need to worry about a knowledge cutoff for this task, because web search is on by default in all three apps and current models' training data runs to within a few months of today.
Leave the mode on its default for a first task. The assistant decides how hard to think; for a bigger job you can raise the effort or thinking level, or switch to a research mode, which the Fluency page Decide covers. Send the brief and give it a minute. You should see it search, read the attachment, and then write.
Checkpoint: The response refers to your attachment and to pages it looked up, not just to general knowledge.
Step 4: Check the result, then redirect or restart
Goal: Decide whether the work is right before you use it.
Action: Read the answer against your brief in three passes. Did it meet the objective and every constraint? Did the verification come back, with a page named for each venue? Then click two of those pages yourself and confirm the hours. Spot-checking the claims that would ruin your weekend matters more than checking everything.
If something is off, say what and why in one follow-up. The assistant keeps the whole conversation in view, so you do not need to repeat the brief.
Saturday is too packed and the second venue is 40 minutes away by bus. Replace it with something within a 15-minute walk of the hotel, and keep Saturday to two activities.
If two corrections in a row have not fixed it, stop correcting. Open a new conversation and send a better brief that includes what you learned. Anthropic's own guidance for its coding agent says the same: a clean session with a better prompt almost always beats a long session full of accumulated fixes.
Checkpoint: You have a result that meets the brief, and you have personally confirmed the two facts that matter most.
Step 5: Save the brief, not just the output
Goal: Make the next similar task faster.
Action: Copy the final output to wherever you will use it. Then copy your brief into a notes app, with a line about what you changed after the first attempt. Treat it as a template with blanks: next time, swap the city, dates, and budget.
Two things to know about what the assistant keeps. Your chat history is saved in the app. Memory is also on in Claude's free plan and in ChatGPT's paid plans (off by default in the EU and UK), so the assistant may carry details across conversations. For anything you would rather it not remember, use an incognito or temporary chat; the Fluency page Protect explains the settings.
Checkpoint: Your output is saved outside the app, and your brief is stored where you will find it.
Common pitfalls
- Vague objective: "Help me with my trip" gets a generic reply. The one-sentence definition of done is the fix.
- Asking for suggestions when you want work: Current models do what you literally asked. Say "plan", "draft", "compare", or "build".
- Skipping verification: It is the most-skipped field and the one that catches errors. Always ask the assistant to show where a fact came from.
- Piling unrelated tasks into one chat: Each task gets its own conversation. Mixed contexts produce muddled results.
- Treating the first answer as final: In Microsoft's 2026 Work Trend Index, 86% of AI users said they treat output as a starting point. Join them.
Key takeaways
- Delegation starts with a task you can define in one sentence
- A six-field brief (objective, context, resources, constraints, output, verification) replaces the old "task, context, format" prompt
- Attach files and let the assistant search; do not retype or paste
- Check the claims that matter, redirect once or twice, then restart with a better brief
- Save the brief as a template; the output is disposable, the brief is reusable
Next steps
Continue to: Use AI as a Research Assistant to run a bigger task in research mode and learn to check citations.
Continue to: Brief: How to Ask AI for What You Want for the full treatment of each field and the examples that steer format.
Continue to: Decide: What to Delegate, Which Tool, Which Mode if you are unsure which tasks to hand over.