10 moves that turn you into an AI Agent Manager this week.

No job title required. Just the stack.

In February 2026, Harvard Business Review published a piece called "To Thrive in the AI Era, Companies Need Agent Managers" by Suraj Srinivasan of Harvard Business School and Vivienne Wei, COO of Salesforce's Agentforce platform. It named a job title that didn't formally exist 12 months earlier.

An agent manager is the person responsible for making sure AI agents deliver actual business outcomes. Not an IT admin. Not a data scientist. Someone who sits between strategy and the autonomous systems doing the work.

Salesforce, JPMorgan, and Walmart are already hiring for it. Average salaries are sitting around $103,000, with experienced operators clearing $175,000. The HBR piece compares the role to product managers in the early 2000s. The Interview Guys put it more bluntly: "Social media manager didn't exist in 2005. By 2015 every mid-sized company had one. We're at the 2008-equivalent moment for AI Agent Manager."

Translation: the window is open right now. You don't need to wait for the job posting. You can build the stack yourself this week and walk into interviews already operating at the level the role requires.

Here are the 10 moves that get you there.

Audit your repetitive tasks.

Before you build anything, you need to know what to build. Open a blank doc. Three columns: Task, Frequency per week, Time per occurrence.

Anything you do 3 or more times a week that follows the same pattern goes on the list. Pick the top 3 by total time cost. Those are your first builds.

Stop at 3. The mistake everyone makes is auditing 20 tasks, getting overwhelmed, and building zero. Three is enough. You can always come back.

These three are the ones that put money in. Build them first.

Wire up a cold email agent.

Give Claude your ICP, your offer, and 3 emails that actually landed replies. Tell it to generate 20 variations. Send the top 5 by your judgment. Track open and reply rates. Feed the winners back. Iterate.

The system prompt that works:

Stack: Claude (or GPT-5) + Smartlead/Instantly for sending + a Google Sheet for tracking.

Build an outreach list agent.

Pair Apollo or Clay with Claude. Feed it your ICP criteria. The agent finds prospects, scores them against your fit framework, and writes a personalised first line for each one based on their LinkedIn, recent posts, or company news.

This is the SDR role, automated. The only thing left for you is reply management, which is the part that actually requires a human anyway.

Build a follow-up agent.

When a lead goes 7 days without replying, you lose them to inertia. Most warm pipelines die here, not at the cold stage.

The fix: Connect your CRM (HubSpot, Pipedrive, Folk, whatever) to Claude via Make or n8n. When a deal goes cold, the agent drafts a personalised nudge using the conversation history, sends it to your inbox for review, and you approve with one click.

You stay in the loop on every send. You just don't have to write them anymore.

Stack: Make.com or n8n (free tier works) + Claude API + your CRM. Setup takes about 90 minutes. Pays for itself in one recovered deal.

These three compound. Build them once, they keep paying.

Build a landing page agent.

Brief in. HTML out. Vercel deploy in under 10 minutes. The full pipeline lives inside Claude Code.

Stop paying designers £300 for pages that take two weeks. Build, test, kill, rebuild fast. The pages that work, scale. The ones that don't, delete and move on.

Pair this with the design skills stack from last week's post (frontend-design, impeccable, 21st.dev) and the output stops looking like AI slop. It looks like Linear's site.

Set up a content repurposing agent.

Record once. Feed the transcript to Claude. Get back a newsletter section, 3 LinkedIn posts, 5 tweet hooks, and an Instagram reel script.

This is the loop that turned a single 12-minute conversation about Claude Code into a reel that did 293,000 plays. Same audio, five different distribution surfaces.

The prompt:

Create an SEO content agent.

Give it a target keyword, your top 3 competitors' ranking pages, and your brand voice samples. Output: a content brief, a first draft, internal link suggestions, and a meta description.

Set this running on 2 posts per week. Watch the compounding start somewhere around month 3. By month 6 you've got an organic moat that costs almost nothing to maintain.

Stack: Claude or Claude Code + Ahrefs/SEMrush for keyword research + your CMS. The agent does 80% of the work. You edit, fact-check, and ship.

These two buy you time and signal. Cheap to build, painful to live without once you have them.

Build a research agent.

Set up Claude or Perplexity to run competitive intel on a schedule. The prompt:

Schedule it via a cron job, GitHub Action, or just a recurring calendar reminder that runs the prompt for you. Twenty minutes to set up. Saves hours of manual research every week and makes you sound like the most prepared person in every meeting.

Set up an analytics digest agent.

Every Monday, the agent pulls your key numbers (Stripe, Fathom, Google Search Console, GA4, whatever you actually use) and writes a plain-English summary in your inbox.

What's up. What's down. What to do about it. No dashboards to log into. No spreadsheets to scroll. Just signal.

Stack: Claude API + Make.com or n8n for the API pulls + a simple email send action. About 90 minutes to wire up. After that, you never miss a signal again.

Build your agent manager dashboard.

This is the move that ties everything together and makes you a manager instead of a user.

One simple page. Notion, Airtable, or a markdown file in a GitHub repo if you want to keep it dev-native. Five columns:

Every Monday, you check the dashboard. Anything red, you triage. Anything green, you trust. Anything new comes up, you add a row.

This is what the HBR piece is actually describing. You're not running tasks anymore. You're running agents. The dashboard is the artefact that makes that real, both for you and for anyone who hires you.

Build the order. This week.

Don't try to build all 10 at once. The order matters more than the speed.

1. Today: Audit (Move 01).
2. This week: Pick ONE income agent (Moves 02, 03, or 04). Build it end-to-end.
3. Next week: Add a growth agent (Moves 05, 06, or 07).
4. Week 3: Add an ops agent (Moves 08 or 09).
5. Week 4: Build the dashboard (Move 10).

In 30 days you have 4 working agents and a dashboard. That's not a stack. That's a job description.

Companies are hiring for this role at $103k to $175k. They're spending months interviewing candidates who can't produce a working agent in the technical screen.

You can have 4 by the end of the month. Bring screenshots. Bring outputs. Bring the dashboard.

Or skip the interview entirely and run them for yourself.