The posting in the reel is Decagon's Agent Strategy Manager in San Francisco. Two things are printed on it. The pay, "$160K-$220K OTE + equity", and this line in the requirements: "Strong technical acumen and passion for the overall agent building process (ability to code not required or needed)." Read the pay properly: OTE means on target earnings, so that figure is base plus commission if you hit target, not a base salary. Decagon also has a second title, Agent Development Manager in New York, at $150K to $220K. Those two are not the same band, they just share the same $220K ceiling. Harvard Business Review named the role in February 2026 in To Thrive in the AI Era, Companies Need Agent Managers.
Every listing in this family asks for the same four abilities, and the four credentials below map onto them. That is the whole logic of the order.
Cost: Microsoft publishes no price for this one, on the credential page or in its Applied Skills FAQ, and the assessment launches without a checkout step. Go in expecting to confirm the cost yourself at launch. Time: a 30 minute hands-on lab. Level: beginner. You get: a credential that appears on your Microsoft Learn profile.
This one goes first because it is hands-on rather than a quiz, and it is over in half an hour. The assessment drops you into a live lab and scores you on three things: deploy a model, create an agent, test the agent. That is the smallest complete loop of the job, and doing it once makes every other item on this list make sense.
Credential page: Get started developing agents in Microsoft Foundry. Microsoft's lab FAQ is where the 30 minutes comes from, since beginner labs get 30 minutes and you can request a little extra time.
Cost: free. Time: 3.5 hours, 13 lessons across 5 sections plus a quiz. Prerequisites: none, and it assumes no technical background and no prior experience with AI tools. You get: a working mental model and a completion badge.
This is the one that turns you from someone who reports the bug into someone who diagnoses it. It teaches the four properties underneath every odd AI answer: next token prediction, knowledge, working memory, and steerability. The point is that by the end you can look at an unexpected output, recognise which kind of unexpected it is, and apply the right fix instead of just rewriting the prompt and hoping.
Course page: AI capabilities and limitations. You do not need a paid Claude account, only a login for the learning platform.
Cost: the trail page prints no price, in either direction. Time: about 10 hours 18 minutes, 23 badges. Updated: 14 September 2026. You get: Agentblazer Champion status.
This is the one I would pick if you only do a single item on this list, and you should know before you start that it is also the longest item here, longer than the other four put together in study time. It ends with a hands-on step called Quick Start: Assemble a Service Agent with Agentforce Builder, where you create a service agent that helps with bookings and recommendations. Finishing it means you can say you have built and assembled a working agent, which is a different sentence to saying you studied agents.
Trail page: Become an Agentblazer Champion 2026. Trailhead is free to use and so are the badges.
Cost: $150 exam fee. Time: 90 minute exam, 70% to pass. Valid: three years. You get: a proctored professional certification.
This is the only one on the list you pay for, and it is last because it assumes you have already done the rest. Its topic list reads like the job description: agentic discovery, agent blueprint design, context grounding, escalations, agentic evaluations, prompt engineering, and agentic orchestration. UiPath says in its own exam description that it "is well suited for individuals in both technical and non-technical roles", and there are no prerequisite exams. UiPath publishes the fee, the duration, the pass mark and the validity, but not a question count, so do not plan your pacing around a number nobody has printed.
Certification overview: UiPath certification. The full exam description, including the fee and the topic breakdown, is in the official exam description PDF.
Four credentials on a profile looks like collecting. One narrative looks like a candidate. Paste this into Claude once you have finished two or more of them.
I am targeting AI agent manager roles. The real job titles are agent strategy manager, agent development manager, agent deployment engineer, AI operations manager, and agentic operations architect. Here is what I have actually completed so far: [list the certificates you have finished and the one concrete thing you built or tested in each].
Here is my background: [two or three sentences about your current job and industry].
Write me three things. First, a four-sentence summary I can put at the top of my CV that leads with what I can do rather than what I completed. Second, a short worked example, six sentences maximum, describing one agent I could realistically run in my own industry: what it would handle, what it would never be allowed to decide, and how I would know it was working. Third, the five questions I should expect in an interview for this role, with a one-paragraph answer for each based only on what I have genuinely done.
Ask me up to three questions first if you need detail to make this specific. Keep it plain, no buzzwords, no em dashes, and no claims I have not earned.This is the artefact that gets you hired, and it is the same document an agent manager writes in week one. Run it on a process you already know well.
Help me write an agent brief for a process I know well, the way an AI agent manager would write it for their team. The process is: [describe it step by step, including who does it today and how long it takes].
Before writing, ask me up to three questions about what a good outcome looks like and what I would reject outright. Then write the brief with these sections: the goal in one sentence, the inputs the agent receives, the steps in order, the exact output format, the quality bar, an explicit list of things it must never do, and the situations where it has to stop and hand the work back to a human instead of guessing.
Finish with an evaluation plan: ten test cases drawn from the process I described, what a pass looks like for each, and how I would report the score to a manager who has never heard of any of this.