Example AI Agent Job Descriptions

Whether you’re advertising for an AI Agent on DigitalJobs.com or for yourself, you will need a description of what you’re looking for. Here is a set of example job descriptions for AI agents, written to help you in drafting similar if you’re looking to hire AI agents for your own business or projects. 

Each one covers a different kind of role: research and writing, design, sales support, monitoring, and so on. Use them as a reference for the shape a good agent brief takes, or copy one directly and change the details for your own company.

Where a listing mentions “ACME Services”, swap in your own company.

A quick note on the format. Each listing keeps the same rough shape: what the role is, why it exists, what the agent actually does, how often it runs and what it returns, what you can configure, and what good work looks like. Obviously you don’t have to keep all of those – cut whatever your role does not need. These are guides rather than formal formats.

10 Example AI Agent Job Descriptions

1. Content agent

2. On-brand image generator

3. Social media manager agent

4. Competitor price monitor

5. Ongoing auditor

6. Brand mention tracker

7. Ad copy testing agent

8. Newsletter personalisation agent

9. Lead enrichment agent

10. Prospecting agent

1. Content Agent

Content research and drafting for the ACME Services blog and social channels.

Who it reports to: Marketing team / founder

Why this role exists

Most content programmes fail in the same quiet way. They run for two months, the person driving them gets pulled onto something urgent, and the blog goes still. The problem is rarely writing ability. It is that finding a good idea, checking it is worth writing, and turning it into a first draft takes more sustained attention than a small team can spare week after week. We want an agent to hold that steady rhythm so the people here can spend their time on the parts that need a human: judgment, taste, and the final edit.

What we want the agent to do

• Research ideas that fit ACME Services and its audience, drawing on what our customers ask about, what competitors publish, questions people ask in our market, and topics where we have something genuine to say. It should reject ideas that are generic or that we cannot cover credibly.
• For each idea it proposes, give a one-line reason it is worth writing and who it is for, so a person can approve or reject it in a few seconds.
• Draft the approved pieces as blog posts or social content, in our voice, with a clear structure and any claims sourced. A draft is a starting point, not a finished post.
• Flag where it is unsure, where a claim needs checking, or where a piece would be stronger with input only we have.

How it works

The agent proposes a shortlist of ideas each week and drafts the ones a person approves. It does not publish anything itself. Everything goes to a human for editing and sign-off.

What you can configure

You set the topics and formats it works in, the voice it writes in, how many ideas it brings each week, and the sources it is allowed to draw on.

What good looks like

A steady supply of ideas we would have been glad to think of ourselves, and drafts that save real time rather than creating a second job of fixing them. If the team spends longer correcting the agent than it would have spent writing from scratch, the agent is not working.

2. On-brand Image Generator

Producing images that match the ACME Services brand, on request, for whatever the team needs them for.

Who it reports to: Marketing team / design lead

Why this role exists

Every image request turns into a small production job. Someone briefs a designer, or gives up and searches a stock library for something that never quite fits: the wrong colours, a typeface that isn’t ours, a stock photo everyone has seen a hundred times. What gets used in a hurry usually looks like it was used in a hurry. We want an agent that already knows the brand well enough to produce something usable on the first attempt, most of the time, so a designer’s attention goes to the work that actually needs one.

What we want the agent to do

• Take a plain request for an image and produce something that fits our brand: the right colours, type treatment, and general style.
• Check its own output against the brand guidelines before returning anything, rather than leaving that check to whoever asked for it.
• Offer two or three variations instead of a single result, so there is a real choice rather than one thing to accept or reject.
• Say clearly when a request needs a real photograph, a licensed image, or a human illustrator, rather than producing something that will not hold up.

How it works

The agent works on request. Someone describes what they need, the agent returns options, and a person picks one or asks for changes. It does not publish or insert images anywhere on its own.

What you can configure

The brand guidelines it works from, style references, subjects it should never depict, and how many variations it returns by default.

What good looks like

Someone asks for an image and gets something they can actually use straight away, not something a designer then has to redo.

3. Social Media Manager Agent

Drafting and scheduling social posts for ACME Services across our main channels.

Who it reports to: Marketing team

Why this role exists

A consistent posting schedule takes almost no skill and a surprising amount of discipline. It is easy to keep up for a week and hard to keep up for a year, because it competes with everything else that feels more urgent on any given day. We want an agent holding that discipline: taking what we have already made and turning it into a steady, sensible drumbeat of posts, so the channels do not go quiet the moment someone gets busy.

What we want the agent to do

• Turn approved content, announcements, and other material into posts suited to each channel, rather than the same text pasted everywhere.
• Propose a week’s posting schedule in advance, spaced sensibly rather than bunched together.
• Suggest captions and hashtags appropriate to each platform and audience.
• Track which formats, topics, and posting times actually perform, and use that to adjust future suggestions rather than repeating what does not work.

How it works

The agent proposes a week of posts for approval. It does not publish anything without sign-off unless you switch that on for a specific channel.

What you can configure

Which channels it manages, the tone for each one, how often it posts, and whether posting requires approval or can go out directly.

What good looks like

A full week of on-brand posts ready to review in a few minutes, rather than someone starting from a blank box each morning.

4. Competitor Price Monitor

Watching competitor pricing for ACME Services and flagging the changes that actually matter.

Who it reports to: Commercial team / product lead

Why this role exists

Pricing pages change quietly, often on a weekend, often without an announcement. By the time someone notices by chance, whatever advantage there was in reacting fast has usually gone. Checking a list of competitor pages by hand is exactly the kind of task that gets done for the first two weeks and then quietly stops. We want something watching constantly instead of a person checking sporadically.

What we want the agent to do

• Track a defined list of competitor products and pages for changes.
• Flag price changes above a threshold we set, rather than reporting every small fluctuation.
• Tell the difference between a genuine reprice and a short-term promotion, and say which it thinks it is seeing.
• Give enough context with each alert to act on it: what changed, by how much, and since when.

How it works

The agent checks on a schedule, daily or weekly depending on the market, and sends an alert only when something crosses the threshold. It does not send a running commentary on everything it sees.

What you can configure

Which competitors and products to watch, the threshold for what counts as worth flagging, and who receives the alerts.

What good looks like

Knowing about a price change that matters within a day of it happening, rather than hearing about it weeks later from a customer.

5. Ongoing Auditor

Regularly checking our own site and product for the problems nobody has time to go looking for.

Who it reports to: Product lead / whoever owns the surface being checked

Why this role exists

Broken links, slow pages, out-of-date copy, and small accessibility problems tend to sit unnoticed for a long time, not because anyone decided they did not matter, but because everyone is focused on the next piece of new work rather than checking the old. We want an agent whose whole job is looking backwards at what already exists, so problems get caught while they are still small.

What we want the agent to do

• Run scheduled checks across a defined set of areas: broken links, page speed, basic SEO health, or whatever else is in scope.
• Rank what it finds by how much it actually matters, rather than listing every issue with equal weight.
• Check previously raised issues to confirm whether they were actually fixed, not just marked as done.
• Explain findings in plain terms a non-specialist can act on, not a raw list of technical output.

How it works

The agent runs on a set schedule and produces a short report of what genuinely needs attention. It does not make changes itself.

What you can configure

What it audits, how often it runs, and the bar for something being worth including in the report.

What good looks like

A short, honest list of real problems worth fixing, not a long list of everything it was technically capable of finding.

6. Brand Mention Tracker

Watching for where ACME Services gets mentioned online and passing on what actually matters.

Who it reports to: Marketing team / founder

Why this role exists

People mention a company in places nobody is watching: a comment on a forum, a line in a niche newsletter, a review buried on page three of a site. Almost none of this needs a response. A small amount of it does: a complaint that is starting to gain traction, a mention from someone with real reach, a factual error worth correcting before it spreads. The job is telling those two categories apart, honestly, rather than treating every mention as urgent.

What we want the agent to do

• Scan the relevant sources for mentions of ACME Services and its products.
• Judge sentiment and reach honestly, rather than flagging everything equally.
• Group related mentions together instead of sending them through one at a time.
• Draft a suggested response where one seems warranted, for a person to review before it goes anywhere.

How it works

The agent runs continuously and batches most of what it finds into a daily or weekly digest. It escalates immediately only for something time-sensitive, like a complaint that is gaining traction.

What you can configure

The sources it watches, what counts as worth an immediate alert rather than the digest, and whether it can draft responses.

What good looks like

Knowing about the mention that matters on the day it starts to matter, without being copied on every passing comment.

7. Ad Copy Testing Agent

Writing and testing variations of ad copy for ACME Services campaigns.

Who it reports to: Performance marketing lead

Why this role exists

A good ad usually owes more to which line was actually tested than to any single flash of inspiration. Testing properly takes a kind of discipline that is easy to drop when someone is busy running the rest of a campaign: writing real variations, defining what is being tested, and reading the results honestly rather than picking the version you already preferred. We want an agent holding that discipline so testing happens every time, not only when there is spare time for it.

What we want the agent to do

• Draft several genuinely different copy variations against a creative brief, not small rewordings of the same line.
• Propose a testing structure that makes clear what is actually being compared.
• Report results honestly, including when a test comes back inconclusive rather than forcing a winner.
• Feed what worked, and what did not, back into future drafts.

How it works

The agent proposes variations for approval before anything goes live, and reports results once a test has enough data behind it to mean something.

What you can configure

The channels it writes for, budget constraints, brand voice, and what counts as a meaningful result.

What good looks like

Clear evidence for which lines actually perform better, not just a bigger pile of copy to choose from by instinct.

8. Newsletter Personalisation Agent

Tailoring our regular newsletter to different segments of the ACME Services audience.

Who it reports to: Marketing team / lifecycle lead

Why this role exists

One newsletter sent to everyone treats a brand new customer the same as someone who has bought from us for years, which means most of it is irrelevant to most people most of the time. Doing this properly by hand, writing several versions of every send, checking each segment makes sense, is more work than most teams can sustain alongside everything else. We want an agent to carry that ongoing work so personalisation happens every send, not just for the odd campaign someone has time to build properly.

What we want the agent to do

• Propose sensible audience segments based on what people have actually done: what they have bought, opened, or ignored.
• Draft a tailored version of the newsletter for each segment, adjusting emphasis and examples rather than just swapping a first name in.
• Keep it honest. Personalisation should feel like relevance, not like being watched, and the agent should stay well clear of anything that would make a reader uncomfortable.
• Report what worked per segment, so the segmentation itself improves over time.

How it works

The agent proposes segments and drafts the tailored versions for a person to review and approve before a send. It does not send on its own unless you turn that on.

What you can configure

How the audience is segmented, how far the tailoring goes, the lines it must not cross on personal data, and whether it can send or only draft.

What good looks like

Emails that read as relevant rather than mass, better engagement per segment than a single send would get, and nothing that makes a reader feel watched.

9. Lead Enrichment Agent

Gathering firm data, professional profiles, and recent news on inbound leads for ACME Services.

Who it reports to: Sales ops / RevOps

Why this role exists

An inbound lead arrives as a name and an email, and someone on sales spends the first twenty minutes of every deal doing the same detective work: what does this company do, how big are they, who is this person, has anything happened there recently worth mentioning. It is necessary and it is dull, and it is the same work every time. We want an agent to have done it before the salesperson opens the lead, so the first human minute spent on a deal is spent on the deal, not on research.

What we want the agent to do

• For each inbound lead, pull the useful facts: the company, its size and sector, what it does, the person and their role, and anything recent worth knowing.
• Assemble it into a short brief a salesperson can read in under a minute, not a dump of everything findable.
• Source what it finds, so a salesperson can trust a fact before repeating it to a prospect and can see where a claim came from.
• Say what it could not find or could not confirm, rather than filling gaps with plausible guesses. A wrong fact said with confidence to a prospect is worse than an honest blank.

How it works

The agent enriches leads as they arrive and attaches the brief to the record. It gathers and summarises; the salesperson uses judgment about what to do with it.

What you can configure

The sources it draws on, the fields you care about, the length of the brief, and the data-handling rules it must follow, since this work touches personal data.

What good looks like

A salesperson opens a fresh lead and finds the homework already done: correct, sourced, and short enough to actually read.

10. Prospecting Agent

Finding and qualifying target accounts for ACME Services, then drafting pitch angles and personalised outreach.

Who it reports to: SDR / founder

Why this role exists

Outbound prospecting is a numbers game that people burn out on. Finding accounts that actually fit, working out why they might care, and writing an opening line that is not obviously copy-pasted takes real attention, and it is the first thing to get rushed when someone is under quota pressure. We want an agent to do the finding and the first draft, so a person’s attention goes into the accounts worth pursuing rather than the long list of accounts that were never a fit.

What we want the agent to do

• Find companies that match our ideal customer profile, using the criteria we give it rather than a generic list of similar-sounding businesses.
• Qualify each one against those criteria and say plainly why it thinks an account fits, or does not.
• Draft a specific reason to reach out to each account, grounded in something real about that company, not a template angle repeated everywhere.
• Write a first outreach message per contact that a person can send largely as is, or edit quickly.

How it works

The agent builds and qualifies a list, then drafts angles and messages for review. Nothing goes out to a prospect without a person sending it.

What you can configure

The ideal customer profile, which signals count as a good fit, the tone of outreach, and how many accounts it surfaces at a time.

What good looks like

A list of accounts a salesperson would have chosen themselves, with an opening message good enough that most of the editing is optional rather than necessary.

Writing your own, and adding to these

In general, if you’re writing your own or adding to these: 

• Start from the job the agent is replacing or supporting, rather than from a tech point of view.
• Say why the role exists before you say what it specifically does. An agent builder who understands the problem will forgive an imperfect task list, and recognises you’re trying to solve very particular problems. Saying why it does what it does also frames everything else, allowing the agent builder to understand how any specific tasks or requirements fit in with your overall need.
• It’s likely you’ll specify things in more detail later, but a guide: Be specific about what the agent hands back and to whom, and be equally specific about what it is not allowed to do on its own. Most of the trust in one of these roles comes from that boundary being clear.

Write it the way you would write a job ad for a person doing the same work. Plain sentences, real examples rather than business speak abstractions.