Welcome to Prompt, Tinker, Innovate—my AI playground. Each edition gives you a hands-on experiment that shows how AI can sharpen your thinking, streamline your process, and power up your creative work.
This week's playground: Run a team of agents in a single session
Scroll back through your last long AI session. You'll probably see the same loop: you ask for something, read the answer, then ask for the next thing. The AI waits for you, and you wait for the AI.
Imagine this: It's 9 p.m., and tomorrow morning you have a call with a company you barely know. You need to know how they describe themselves, who runs what, and what's changed lately. That's three research jobs. The usual way, you or your AI work through them one by one.
Agents let you run those jobs side by side. In one prompt, you can ask Claude or ChatGPT to assign each job to a different agent, then have a fourth challenge what the others found. The assistant then pulls everything into one briefing before your call.
Not every task needs a team. A quick question, or work where each step builds on the last, still belongs in a single prompt. When the parts don't depend on each other, agents can speed up the work and catch more mistakes, especially when one agent reviews the others' work.
Use case spotlight: where agents fit
Company research is this week's experiment, but the pattern works elsewhere:
- Competitive research: Agents cover a competitor's positioning, pricing, and customer feedback at the same time. A final agent flags claims that lack a source.
- Content team: Agents draft a LinkedIn post, an email version, and subject lines. An editor checks them against your style rules.
- Audience panel: Agents take the perspectives of a CFO, a practitioner, and a skeptic, then react to the same pitch.
- Document review: Agents each read a section of a long report. A final agent looks for conflicts across their summaries.
The test is simple: can the parts be done independently, or would a second set of eyes help? If so, try a team of agents. If not, such as a short email you need to send, one prompt is faster.
Your AI experiment: Try this prompt
👉 Time to tinker: Pick a competitor and replace the brackets below. In ChatGPT, choose Work mode. In Claude, choose Cowork if you see a Chat/Cowork switch; otherwise, start in the unified Claude interface.
📝 Prompt:
Goal: Create a one-page briefing on [COMPANY] to prepare me for [a sales call / a partner meeting / an interview]. My context: [one sentence on what I sell, offer, or want from this meeting]. Use it to decide which leadership, priorities, and news matter most. Report what you find even if it hurts my case. Step 1. Assign three agents to work in parallel: - Positioning Analyst: Read the company's homepage, product pages, and last 3 blog posts. Report its core claim, target buyer, and the 3 phrases it repeats most. - Leadership and Priorities Researcher: Find the company's leadership team, any leadership changes in the past 12 months, and its stated priorities from its website, annual report, or careers page. Report who runs what and the 3 priorities that show up most. Flag anything you can't confirm. - News Researcher: Find the company's announcements, press releases, and news coverage from the past 6 months. Report the 3 most important developments, with dates. Rules for every agent: Keep your report under 250 words. Link every source. Label each finding "confirmed" or "inferred." If you can't open a source, say so. Don't fill the gap. Step 2. After all three reports are ready, assign a fourth agent called the Skeptic. Give the Skeptic all three reports. Have it flag claims without supporting sources, findings that contradict one another, and any finding that looks chosen to fit my context instead of what the company is doing. Step 3. After the Skeptic reports back, create the final briefing: an executive summary, what the company does well, where its story has gaps, and 3 questions I should ask in the meeting, based on my context. Include what the Skeptic challenged and how you resolved it. Show me the agent list as the agents run.
💡 Pro tips: Before you hit enter
- Keep the checker separate from the writer. The Skeptic reads the three research reports and flags weak claims. The assistant then writes the briefing, using those flags. A reviewer that didn't produce the work has no reason to protect it.
- Give every agent the same report format. Ask for the same headings and a word cap, like the 250-word limit in the prompt. The assistant can compare the reports side by side, and no single long report dominates the briefing.
- Watch your token usage. Four agents each search, read pages, and write a report, so this prompt uses more of your limit than a single prompt would. Save it for work that justifies the cost. Check your usage after the first run, and cut the number of agents if you're on a tight plan.
What did you discover?
Think about your own week. Which tasks have parts that don't depend on each other, or would benefit from a second agent checking the first one's work? Those are the ones to hand to a team. Which would you keep as a single prompt, because splitting them adds cost and steps without a better result?
Tell me in the comments which task you'd split, and which you'd leave alone.
Until next time—keep tinkering, keep prompting, keep innovating.
📩 Not subscribed yet? Hit the button at the top to get the next experiment.
