Agents

The 5-Agent AI Crew Blueprint

7 minute readUpdated June 2026Explore more

TL;DR

A crew is not one prompt doing everything - it is five agents that each have one job and hand their output to the next. The Research Agent finds what your audience is talking about, the Content Agent turns it into a post in your voice, the Lead Qualifier ranks who is worth your time, the Follow-Up Agent writes the message, and the Performance Agent tells you what to do more of. Copy the five prompts below and you can run the whole crew by hand in an afternoon.

What a crew actually is

Most people think AI automation means prompting one model and hoping for a decent draft. A crew is different. Each agent has one job and a clear handoff. The Research Agent finds what is trending. The Content Agent turns that into a post. The Lead Qualifier decides who is worth following up with. The Follow-Up Agent writes the message. The Performance Agent tells you what is working. You set the rules once and let them execute.

Agent 1: The Research Agent

Job: find what your audience is already talking about so your content lands instead of disappearing. It scans the places your audience gathers for pain points, questions, and trending frustrations, and hands you a list of content angles. No more staring at a blank screen.

You are a content research agent for [YOUR NAME], a [YOUR ROLE] who helps [YOUR TARGET AUDIENCE] achieve [THEIR OUTCOME].

Your job: scan the following sources and identify the top 10 pain points, frustrations, or questions my audience is expressing right now.

Sources to analyze:
- Reddit: r/[YOUR NICHE SUBREDDIT]
- Comments on recent posts from [3 COMPETITORS OR PEERS]
- Any customer messages or DMs I paste below

For each pain point, output:
1. The exact language people are using (quote where possible)
2. The emotion underneath it (fear, frustration, embarrassment, etc.)
3. One content angle that addresses it directly

Do not editorialize. Use their words, not marketing language.

Customer messages to analyze: [PASTE HERE]

Agent 2: The Content Agent

Job: turn research output into ready-to-post content in your voice. Feed it the pain points from Agent 1 plus a few examples of your best past content. It learns your tone, sentence length, and proof-point style, then drafts posts that sound like you - not like a robot pretending to be you. Give it feedback for two weeks and the drafts get sharp.

You are a content writing agent trained on [YOUR NAME]'s voice. Study the writing samples below carefully before drafting anything.

Voice rules extracted from the samples:
- [LET THE MODEL IDENTIFY YOUR PATTERNS, THEN PASTE THEM BACK HERE FOR NEXT TIME]

Writing samples: [PASTE 3-5 OF YOUR BEST POSTS HERE]

Now write a post using this pain point from my Research Agent: [PASTE PAIN POINT]

Post structure:
- Line 1: transformation or big-number hook (no questions, no fluff)
- Line 2: "Here is how:" or "Here is what I found:"
- Body: 5-8 bullet points or numbered steps, each concrete and actionable
- Closing line: one sentence telling them what to do next

Target length: 150-200 words. No hashtags. No em dashes. No bold text.

Agent 3: The Lead Qualifier

Job: look at everyone who comments, likes, or DMs you and tell you which ones are worth following up with today. Ignoring every comment is wrong; trying to answer all of them is also wrong. The people ready to buy give signals - specific questions, mentions of a problem you solve, language that shows budget and urgency. This agent reads those signals and returns a ranked list: hot, warm, cold.

You are a lead qualification agent for [YOUR NAME]'s business. We sell [YOUR OFFER] to [YOUR IDEAL CLIENT DESCRIPTION].

Ideal client signals:
- Mentions a specific problem we solve
- Has been in business at least 1 year
- References budget, team, or revenue
- Asks a question that shows they are close to buying
- Uses language like "how do I" or "I have been struggling with"

Disqualifying signals:
- Job seekers
- Competitors doing research
- People looking for free advice with no intent to pay
- Vague comments with no specific context

Here are the comments and DMs from today: [PASTE HERE]

For each one, output:
- Name or handle
- Rating: HOT / WARM / COLD
- One sentence explaining why
- Suggested next step (reply, DM, ignore)

Do not overthink it. Use the signals. Be direct.

Agent 4: The Follow-Up Agent

Job: write the DM or email follow-up for every hot and warm lead the Qualifier flags. This is where most people leak money - a warm lead comes in, life gets busy, the follow-up never happens, and the lead goes cold. Paste in the lead profile and the context; the agent writes a short message specific to what the lead said, ending with one clear next step. Match the message type to where the lead is.

  • Comment reply - within 2 hours of a hot comment. Goal: start a conversation, not a pitch.
  • DM opener - after qualifying a warm lead. Goal: book a call or share a resource.
  • Re-engage message - a lead went cold after initial contact. Goal: reopen the door with new value.
  • Post-call follow-up - within 24 hours of a sales call. Goal: confirm next steps and reduce drop-off.

Agent 5: The Performance Agent

Job: look at your last 10 posts, tell you what is working, and tell you what to do more of next week. Most people post and hope. Paste in your post data - impressions, comments, shares, clicks - and this agent returns three things: what is working, what is not, and the one change to make this week. Run it every Monday morning before the Research Agent so the loop closes on itself.

Wire the five together in one afternoon

  1. 1Open a dedicated workspace for the crew - one conversation or file per agent, labelled clearly.
  2. 2Set up the Research Agent first. Run it Monday mornings; it feeds everything else.
  3. 3Set up the Content Agent second. Paste your best posts as training samples, then run it after Research.
  4. 4Set up the Lead Qualifier third. Run it once a day after checking notifications.
  5. 5Set up the Follow-Up Agent fourth. Run it the moment the Qualifier flags hot leads.
  6. 6Set up the Performance Agent last. Run it every Monday before Research to close the loop.

Common questions

  • Do I need automation software to run this crew?

    No. You can run all five agents by hand inside Claude - one conversation per agent - and get most of the value immediately. Automation (scheduled tasks, cron jobs) is a later step you add once each agent has earned its place. Start manual.

  • Can I run this in Claude Code specifically?

    Yes. Each agent is just a prompt, so it runs anywhere Claude does. In Claude Code you can save each one as a skill or a saved command, point them at real files in your project, and even schedule the recurring ones - see our cron autopilot guide for that.

  • How long does setup take?

    About 90 minutes on day one to get all five running, then 30 to 45 minutes a day to operate them. The Research Agent alone takes about 15 minutes and immediately shows you what to post this week, so start there.

  • Will the content actually sound like me?

    Not on the first draft. The Content Agent learns your voice from the samples you paste and the feedback you give it. After about two weeks of corrections it produces first drafts you only need to edit lightly. The quality is a function of how much feedback you give it early.

  • What if I am not technical at all?

    That is who this is built for. There is no code. You copy a prompt, replace the bracketed blanks with details about your business, and paste in real data. If you can fill in a form, you can run the crew.

  • Which agent should I deploy first?

    The Research Agent. It takes 15 minutes, needs no training samples, and its output becomes your content calendar for the next 7 days. Add one more agent each day until all five are running.

Want help wiring your agent crew together?

Get the other 1 in the agent crew stack - free, with 5,000+ builders.

Join the Club