Running a B2B marketing team and being told to get “agents”? An AI marketing agent takes a job, like turning a webinar into a week of LinkedIn posts, does it with your tools and hands it back for a person to sign off.
TCS’s social team already works this way. They get 80+ posts a month out of Compass Agents, the agent platform we built, and save more than 60 hours of production time a month (TCS case study ).
I wrote the first version of this page in November 2024, and said every marketer would soon have “a full team of autonomous AI agents”, basically their own department. Half right. The agents turned up. “Autonomous” was the wrong word (I don’t use it now), and nobody got a department without doing a lot of writing down first. So this version is about what an agent does in a normal week, and what your team has to give it before it’s any use.
What is an AI marketing agent?
An AI marketing agent is software, built on a large language model, that takes a marketing job, works out the steps, uses your tools to carry them out and checks back with a person at set points.
Anthropic, who make Claude, draw a useful line here. Workflows are systems where the AI follows “predefined code paths”. Agents are systems that “dynamically direct their own processes and tool usage”, and they “can then pause for human feedback at checkpoints or when encountering blockers” (Anthropic ).
In marketing terms, a workflow is “every new webinar becomes three LinkedIn posts”. An agent is “here’s the webinar, work out what’s worth posting, draft it in our voice and show me”.
They’re spreading fast. Microsoft’s 2026 Work Trend Index says the number of active agents in the Microsoft 365 ecosystem grew 15x in a year, and 18x in large enterprises (Microsoft ).
How is an AI marketing agent different from marketing automation?
Marketing automation runs rules a person wrote in advance. If someone downloads the guide, send email two three days later. It does exactly what it’s told, every time, and it can’t cope with anything the rules didn’t expect. An AI marketing agent gets a goal and some limits instead of a rule, for example “turn this webinar into a week of LinkedIn posts in our voice, and flag any customer we can’t name”. It reads the material, decides what to do, uses tools to do it and hands the result to a person to approve. An agent copes with messy material that would break an automation rule, and it can also get things wrong in ways a rule never would, so someone checks its work. Most B2B teams end up with both. The automation platform keeps sending the nurture emails, and the agent drafts the next batch for a person to approve before they’re loaded in.
What can an AI marketing agent do for a B2B team day to day?
The jobs agents do well right now have lots of material going in and a clear standard for what good looks like coming out. That’s drafting, repurposing, research and checking. Deciding what to say, and who to say it to, is still your job.
TCS is the clearest example we’ve got. TCS’s CMO, Ashish Babu, wanted to see what AI could do for the efficiency and creativity of their content marketing, and their Head of Social, Suhail Adam, asked us to help his team with content creation. We set the team up on Compass Agents and trained it on the TCS brand and tone of voice. Now they use it for social posts and for turning video into copy quickly. A newer feature spots the people and companies mentioned in a post and fills in their profile links, the kind of fiddly job nobody misses.
The work splits roughly like this.
| Job | What the agent does | What a person still does |
|---|---|---|
| Social and content drafts | Drafts posts in your voice from a brief, a blog or a recording | Picks what’s worth saying and approves the words |
| Repurposing | Turns a video, webinar or report into copy for each channel | Checks nothing got lost or made up |
| Research | Pulls together what’s public about an account or a topic | Decides what matters, and checks it’s true |
| Checking | Tests content against your brand, tone and SEO rules at scale | Decides which failures matter |
| Campaign admin | Plans campaigns, monitors performance and carries out tasks in tools such as Asana, Notion and Google Analytics | Owns the plan and the money |
AI can check work at a scale no team could. The Partner Benchmarking Tool, an AI platform we built for Microsoft, ran over 40 tests across partner content, checking spelling, grammar, the complexity of the language and tone (Microsoft case study ). No team of people was ever going to read that much partner content line by line. The admin row is what Compass does, the AI marketing platform on our Labs page. It lives inside Microsoft Teams, plans campaigns, monitors performance and carries out tasks through tools such as Asana, Notion and Google Analytics (Labs ).
What agents are still bad at is knowing when something is a bad idea. An agent will happily write a brilliant post about the wrong thing.
What does an AI marketing agent need from your team first?
It needs your rules written down, some examples of good work and a clear line on who signs off what. Without them you get fluent copy that sounds like everyone else’s.
1. Your brand and tone rules, in words
Most brand guidelines are a PDF about logo spacing. An agent needs the words. How you sound, what you never say, which product names need a trademark and which customers you’re allowed to name. For TCS, we trained Compass Agents on the TCS brand and tone of voice before the team got going.
I do the same for myself. There’s a voice file that says how I write and which phrases I never use, and we keep a page listing which clients we can name in public, which every agent reads before it drafts anything. We wrote up how to make your own in how to build an AI writing style that follows you everywhere .
2. Examples of work you were proud of
Compass Agents is trained on the best outputs for each social channel, as well as on TCS’s rules. If you’re setting up your own agent, give it the ten posts you were proudest of and the ten that flopped, and tell it which is which.
3. A sign-off rule
Decide who approves what before the agent writes a word. Ours is simple for anything public. Nothing goes out until a person has approved the exact words. Two more rules came from things going wrong. No agent gets to say a job is done without showing the output, and a second agent attacks every finding before we keep it.
Even then things slip through. A few weeks ago an automated pass on our own website changed “artifacts” to “artefacts” inside a link and broke it, then rewrote an article’s description to say the opposite of the article. Both looked fine at a glance. Checking is a lot of my job now. Did not see that coming.
The heaviest AI users take checking seriously too. In Microsoft’s 2026 Work Trend Index, the people it calls Frontier Professionals, its most advanced AI users, were much more likely to say their team discusses quality standards for AI-assisted work, 54% against 29% for other AI users. It also asks the questions every team running agents has to answer: “Who reviews agent performance? Who has the authority to update the workflows that agents run?” (Microsoft ).
Got a job that eats your team's week?
Tell us what it is. We'll tell you straight whether an agent can do it, and what it would need from you first.
Talk to usShould you build an AI marketing agent or buy one?
Buy one when the job is generic and lives inside a tool you already pay for. Build one when the job depends on your brand, your data and several tools at once, because that’s where an off-the-shelf agent stops knowing enough to be useful.
Most of the big platforms now ship their own agents (I put the main ones on one page in AI agent orchestration: who owns the plan? ). They work well on the data inside their own platform. They know much less about a job that runs across your CRM, your content and your brand rules at once.
Also check you need an agent at all. Anthropic’s advice is to find “the simplest solution possible”, and “This might mean not building agentic systems at all” (Anthropic ). I agree. A prompt and a schedule will do a surprising number of marketing jobs.
Building your own has got much cheaper, though. In the last week of September my AI tools built me 20 tiny tools, each for one job, and most were thrown away within a day or two (is SaaS dead? ). A marketing agent is a bigger build than that, but it isn’t the year-long IT project it would have been two years ago.
Enjoying this article?
Get more B2B marketing insights delivered straight to your inbox.
Thanks for subscribing!
What running agents has changed for me
I write less and check more. Our own marketing is drafted by a handful of agents now. They research, draft and check, and one of them has its own Teams account so it can ask a colleague a question instead of guessing. My part is the brief and the yes or no.
I did get one thing right in the 2024 version. I wrote that the “doers” would become the overseers. Microsoft’s research put a name on it a few months later, the “agent boss”: “someone who builds, delegates to, and manages agents to amplify their impact” (Microsoft ).
Sound is off. Press M or the speaker to turn it on.
So what can an AI marketing agent actually do for a B2B marketing team?
An AI marketing agent can take a big share of the production work off a B2B marketing team, from first drafts to checking, using your tools and stopping for a person to sign off. TCS’s social team gets 80+ posts a month out of one.
It needs your brand and tone rules, examples of good work and a sign-off rule before it’s any use, and deciding what’s worth saying stays with your team. If there’s a job you’d hand to an agent tomorrow, tell us what it is .