Which jobs can you hand to an AI agent, and how do you hire one?

To hire an AI agent, treat it like any hire for a defined job: scope one job, write a brief, run a trial on your own work, accept on evidence, then supervise. In a 2024 article, Anthropic described agents as emerging in production and discussed customer support and coding applications; that is a vendor's account. Whether an agent can take on a job of yours depends on which tasks it can do and which work a person must check. Before you hire AI agents, this section helps you answer that by job, not by tool. For each job it covers what an agent can do today, what you have to give it, and how to hire one without getting burned; where both sides can be sourced, it also compares cost with a person's wage. Every page is researched from cited sources, dated, and lists what we could not verify.

What this section is

This section starts from your side of the hire: whether the job you have in mind is one an agent can do, and how you would know it did it right.

These pages start from the job. A job page breaks the work into tasks and marks each one as done by an agent, done by an agent with a person approving, or kept by a person, with the evidence for each call. It gives a readiness verdict, a brief you can copy, the risks to watch, and, where both sides can be sourced, a cost comparison with a person's wage.

Occupation pages take a human role, such as a bookkeeper, and go through its tasks the same way. Guides cover the steps that apply to every hire.

We do not rank vendors or recommend products, and we have not tested any. A product appears only as evidence, with a source, that a job is being done by agents.

Functions

Pick the part of the business the job sits in. Each hub lists its jobs and roles.

• Function
  AI agents for finance: jobs they do and how to hire one

  AI agents now take on repetitive finance work, such as checking expenses against policy and asking for and using missing context on bank transactions, while people keep approvals, filings and sign-off.
• Function
  AI agents for research: jobs they do and how to hire one

  AI research agents plan a search, read many sources and return a report with citations. Audits show many cited statements are not supported by their sources, so a person checks the citations and owns the conclusions.
• Function
  AI agents for data analysis and BI: what to hand over

  AI agents for data analysis write SQL, build charts and flag broken pipelines from plain-language requests. Their mistakes look like valid answers, so you give them read-only access, a defined metric layer and a known number to check against.
• Function
  AI agents for customer service: what to hand over first

  Help desk vendors document AI agents that answer customer questions using your content (Intercom's Fin) and classify incoming tickets (Zendesk), while people keep refunds, exceptions and upset customers.
• Function
  AI agents for sales: which jobs to hand over first

  CRM and data vendors document AI agents that research accounts, write and send prospecting email, and flag duplicate CRM records, while people keep qualification calls, negotiation and the decision about who gets emailed.
• Function
  AI agents for content creation and marketing: how to hire

  AI agents for content creation can draft page and blog sections, titles, meta descriptions and social posts inside a content editor, according to HubSpot's documentation. Google asks publishers to fact-check all AI-generated content, so a person reviews every piece before it goes live.
• Function
  AI agents for business operations: what to hand over

  Document services such as Azure Document Intelligence extract text, tables and typed fields that agents and applications can use, and Copilot in Outlook recommends meeting times, according to their documentation. Microsoft's guidance for its document service says to keep a human in the loop.

How to hire

Hiring an agent works like any hire for a defined job: scope one job, write a brief, run a trial on your own work, accept on evidence, then supervise.

First, check that the job needs an agent at all. Anthropic, which builds AI models, recommends "finding the simplest solution possible" and notes that this "might mean not building agentic systems at all." OpenAI's guide to building agents names three signs that a job suits an agent rather than fixed rules: complex decisions that need judgment, rulesets that have become hard to maintain, and heavy reliance on unstructured data such as documents or conversation. Both are vendor sources; read them as the builders' own advice.

Second, write down what "done" looks like. The brief below has the same fields as every job page: a goal stated as an outcome, inputs and access, the deliverable, acceptance criteria you can check, volume, systems, constraints and the points where a person approves.

Third, limit what the agent can do. The OWASP Gen AI Security Project lists Excessive Agency (LLM06:2025) as a risk for AI applications and traces it to excessive functionality, excessive permissions and excessive autonomy. Its mitigation for the last is to "require a human to approve high-impact actions before they are taken." In practice, payments, messages to customers and filings are the actions to keep behind an approval. That list is our recommendation, not OWASP's.

Fourth, test before you trust. Anthropic says agents' autonomy brings "higher costs, and the potential for compounding errors", and recommends "extensive testing in sandboxed environments." For a buyer, that can be last month's work with a known answer key, reviewed item by item. OpenAI's guide describes agents that, on failure, can "halt execution and transfer control back to the user"; check that the agent you hire hands off to a named person.

The guides go through each step in detail:

The job board is coming

We are building a job board where you can post a job for an AI agent and the people who build and operate agents can see it. It is not open yet. Leave your email and you will get one message when it opens. If you already have a job in mind, add one line about it; that tells us which jobs to cover next.

If you build or operate AI agents, use the link below the form to tell us which jobs your agent does.

Sources

1. 1
   Building effective agents

   AnthropicVendorAccessed Nov 14, 2026

2. 2
   A practical guide to building agents

   OpenAIVendorAccessed Nov 14, 2026

3. 3
   LLM06:2025 Excessive Agency

   OWASP Gen AI Security ProjectPrimaryAccessed Nov 14, 2026