← Back to all free guides
A free guide by Mariah Brunner ↗ INSTAGRAM ↗ TIKTOK

The Fastest-Growing
Job of 2026

Harvard Business Review says companies running AI agents need a new kind of leader, the agent manager. Knowing your actual job well matters more than being technical. Here’s how to position yourself starting this week.

In February 2026, HBR published "To Thrive in the AI Era, Companies Need Agent Managers" by Suraj Srinivasan, a Harvard Business School professor, and Vivienne Wei of Salesforce. Their argument is simple. Once AI agents do real work, someone has to manage them. They compare it to how product managers showed up during the software boom.

Think project manager, but for AI. You decide what work agents handle, set them up, check their quality, fix what breaks and report the impact. Salesforce already has people in this seat, like the support agent manager HBR profiles, who says "I start and end my day in dashboards, scorecards, and agent observability monitoring."

Why It Matters

Gartner predicted in June 2025 that over 40% of agentic AI projects will be canceled by the end of 2027, because of rising costs, unclear business value or weak risk controls. The person who can prove an agent works and is worth it is the person companies need.

What The Job Actually Is

You don’t need to code or have a computer science degree. Day to day, an agent manager does five things.

01

Pick the work

Look at your team’s workflows and find what’s repetitive and pattern-based enough for an agent. The skill is knowing what should stay human.

02

Set up the agent

Write the instructions, choose what data it can touch, decide when it runs and test it until it’s reliable.

03

Watch quality

Track whether it gets things right, how often it escalates to a person and how fast it works. Catch problems before customers do.

04

Own the handoffs

Decide when the agent stops and hands off to a human, and make that handoff smooth.

05

Report the impact

Show leadership the time and money saved in terms the business cares about.

HBR’s companion tip lists six capabilities for the role. The first two are building "AI operational literacy," meaning you understand how agents work, how prompts shape what they do and how to diagnose failures, and developing deep functional expertise in the work itself. That second one is the good news. If you know your job cold, you have half of it.

5 Things To Do This Week

01

Audit your own job

List every task you do in a typical week and sort each into AI could do it, AI could help or human only. It’s the first thing an agent manager does on the job, and most people find hours they could hand off.

Audit My Week For AI
You are an operations consultant helping me find which parts of my job AI agents could take over. This is the first thing an AI agent manager does, and I'm practicing on my own role. My role: [JOB TITLE AND WHAT I'M RESPONSIBLE FOR] My typical week: [LIST EVERY TASK YOU CAN THINK OF, WITH ROUGH HOURS EACH. ROUGH IS FINE.] Tools I use: [e.g. Gmail, Slack, Salesforce, Excel, Notion] Step 1: Ask me up to 5 questions to fill gaps in my task list (things people forget, like status updates, follow-ups and data cleanup). Step 2: Put every task in a table with these columns: Task, Hours per week, Category, Why, First step. Category is one of: - AI could do this (repetitive, follows a pattern, low stakes if it's slightly off) - AI could help (needs my judgment, but AI could draft, research or organize) - Human only (relationships, sensitive calls, real creative judgment) Step 3: Pick the 3 best candidates for a first agent. Rank them by hours saved and how easy they'd be to check. Step 4: Total the hours in "AI could do this" and "AI could help." Rules: be realistic, not hype. If a task touches private customer data or money, say what guardrail it would need. Don't invent tasks I didn't mention.

02

Run your first agent

Don’t just chat with AI. Manage it. On a paid Claude plan, describe a recurring task in a chat, like "every weekday at 8am, summarize my unread email and flag anything urgent," and click Schedule on the plan it shows you. Run it for a week, note what it gets right and wrong, and fix the instructions. That test, run, fix loop is the core of the job. (How scheduled tasks work)

03

Write instructions like a manager

Most people write requests, like "Can you summarize this?" A manager writes an SOP. "You are the weekly report agent. Every Friday at 3pm, pull completed tasks from the Projects database, group them by project, flag anything overdue and write a summary with these 5 sections." Spell out the trigger, data source, output format, edge cases and rules.

04

Document one process

Take the clearest "AI could do it" task from your audit and write it up as if you were handing it to a new hire, including when it should stop and ask a human. Don’t automate it yet. This document is what an agent manager creates before deploying anything, and it’s a great portfolio piece for an interview.

Turn My Task Into An Agent SOP
You are helping me write a standard operating procedure for a task I want to hand to an AI agent. Write it like I'm handing it to a smart new hire who has never worked here. The task: [PICK ONE FROM YOUR AUDIT, e.g. "the Friday client status email"] How I do it today: [WALK THROUGH IT STEP BY STEP IN YOUR OWN WORDS] A good example of the finished output: [PASTE ONE] Ask me any questions you need first, one at a time. Then write the SOP with these sections: 1. Trigger: what starts it and when 2. Inputs: exactly what data it needs and where it lives 3. Steps: numbered, specific, no vague verbs like "handle" or "manage" 4. Output: the exact format, length and where it goes 5. What good looks like: 3-5 checks a reviewer can tick off 6. Edge cases: what to do when data is missing, late or weird 7. Escalate to a human when: a clear list of stop-and-ask situations 8. Agent instructions: the same SOP rewritten as instructions I can paste into Claude, a skill or a scheduled task Rules: don't add steps I didn't describe without marking them as "suggested." Keep the agent instructions under 400 words.

05

Track your results

Log every meaningful AI task this week in a spreadsheet or a note. Write down what you asked, the time it would have taken by hand versus with AI, and whether the output was usable. On Friday, turn it into a report. "12 tasks that would have taken 8 hours took 2, and 75% were usable as-is" is the kind of line that gets you noticed.

My Weekly AI ROI Report
You are my analyst. Here is my AI log for the week, one line per task: [PASTE YOUR LOG: what I asked AI to do, minutes it would have taken me by hand, minutes it took with AI, and whether the output was usable as-is, needed light edits or needed a redo] Turn it into a one-page ROI report I could show my manager: 1. Headline: total hours saved this week, in one sentence 2. A table: Task, Time by hand, Time with AI, Saved, Quality 3. Hit rate: what percent of outputs were usable as-is or with light edits 4. Where AI struggled and the most likely reason (unclear instructions, missing data, wrong task for AI) 5. One change to my instructions that would fix the biggest miss 6. What I should hand to AI next week Rules: use only my numbers. Do not round up, estimate or add tasks. If a line is missing a number, list it under "Incomplete" instead of guessing. Plain words, no hype.

The Bigger Picture

You don’t need permission or a new title. Start managing AI in your current job and document the results. Either you become the obvious choice when your company creates this role, or you have the portfolio that gets you hired somewhere that already has.