How AI-native marketing teams put AI agents to work
Anthropic published a look at how its own marketing operations team runs on agents, and it is one of the clearest worked examples of an AI-native marketing function published so far. One person on the team used to spend one to two days a week assembling the weekly metrics review. An agent now does most of that in about two hours. The time saved is real. The more useful lesson is what the agent was asked to do: run the whole workflow, from the data hunt to a drafted set of focus areas, on a schedule, with no one clicking through the steps.
Anthropic's marketing operations team turned weekly reporting and campaign builds over to agents. Here is what they changed, and what a CMO should copy.
Short answer: An AI-native marketing team hands whole workflows to agents. Anthropic's marketing operations team runs its weekly metrics review on a scheduled agent that reads meeting transcripts, checks Slack, and queries the data warehouse, cutting one to two days of work to about two hours. What stays with the marketer is validation, enablement, and the data itself.
We build and run AI-native workflows for go-to-market teams, and Anthropic's account matches what we see. Two things decide whether this works: automate the workflow end to end, and keep a person on validation.
What does an AI-native marketing team do differently?
An AI-native marketing team automates the whole workflow. The ordinary version of AI in marketing speeds up one task inside a process a person still runs: a faster first draft, a quicker chart, an AI summary of a dashboard. The AI-native version gives the agent the entire loop and a place to leave the result.
That difference decides the payoff. An assistant that writes a better campaign email saves a marketer minutes. An agent that assembles the entire campaign across your systems saves days. Anthropic is running both kinds of workflow in its own marketing function, and has walked through the setup on a webinar. The account is specific enough to copy. If you want the underlying distinction before you spend on it, we wrote the operating-model version here: an assistant speeds up a task a person owns, and an agent owns the task.
How does the weekly reporting agent work?
The reporting agent runs on a schedule and does the data hunt before anyone asks. Ian Chan, on the marketing operations team, used to spend one to two days a week pulling the weekly metrics review together.
Now a task runs every Sunday evening. The agent reads the previous week's review and the latest meeting transcript, checks Slack for what the sales team is focused on, queries the data warehouse, and leaves a folder with the numbers and a few suggested focus areas. By Monday the review has prepped itself, and the work that used to take one to two days takes about two hours. Anthropic wrote up the same workflow as a tutorial, so the setup is not a secret.
Read that workflow again as a design. The agent runs to a schedule, gathers its own context from three separate systems, does the analysis, and drops a deliverable in a known place. Every one of those is a deliberate choice a marketing leader can make about any recurring report, and none of it depends on the model being clever. It depends on the workflow being drawn that way on purpose.
How does the campaign-build agent work?
The campaign agent runs the setup sequence across every system, so a launch that used to take days takes hours. Annabel Custer, who focuses on campaign operations, used to set up each new event by clicking through Salesforce, HubSpot, Swoogo, and email tools in sequence. She moved that sequence into a Cowork workflow.
The agent now runs the steps across those systems. The manual version was a person carrying data by hand from one tool to the next. The agent version is the same sequence, executed by software that can reach each system. Nothing about the campaign itself changed. The labor of assembling it moved off the marketer's desk.
What changes for the marketers themselves?
The job moves from operating the systems to owning the data and the judgment. When an agent runs the workflow, the marketer stops being the person who clicks through Salesforce and becomes the person who checks the agent's output and answers for the numbers underneath it.
Anthropic describes the shift plainly: Ian and Annabel now spend less time moving through systems and more time on enablement, validation, and the data and processes the marketing team relies on, as more people across the company pull their own numbers and run their own programs. The headcount does not vanish. The work moves up a level, to validating what the agent produced and maintaining the data it draws on. That is the staffing change every AI-native workflow brings, and marketing is no exception. We covered the roles that change on an AI-native team and what an AI-native go-to-market org actually looks like.
Redrawing your own marketing workflows this quarter? Get the next piece in your inbox.
Why do most marketing teams get less from AI than this?
Most teams bolt AI onto the task and leave the workflow untouched. Marketing is one of the functions with the highest AI adoption. McKinsey's State of AI puts marketing and sales among the business areas where companies most often report using it. Yet a lot of that adoption sits at the task level: a copywriting assistant here, a dashboard summary there, each one speeding up a step while the end-to-end process stays as manual as it ever was.
The Anthropic examples work because they start from the workflow. The team started from a sharper question than how to write the weekly review faster: what would it take for the review to prep itself? That question points straight at the workflow, and the workflow is where the payoff lives. Bolting a copywriter onto a process that still needs a human to run every step is why adding AI to an existing workflow so often disappoints.
The capability is arriving regardless of whether teams plan for it. Gartner expects 40% of enterprise applications to ship with task-specific agents by the end of 2026, up from under 5% in 2025. The agents are coming into the marketing stack either way. What a CMO still controls is the design of the workflows they run.
How should a CMO start?
Start with one recurring, data-heavy workflow and give the agent the connections it needs to run it end to end. Anthropic's team did not pick weekly reporting by accident. Recurring work that spends most of its hours gathering data from several systems is exactly what an agent handles well. The sequence that makes it work is short.
- Pick a whole workflow. Choose a recurring output, the weekly review or the campaign setup, that burns hours in data-gathering across systems. Those are the ones that pay back the setup cost. More on choosing the first workflows.
- Give the agent the connections. Anthropic's reporting agent reaches the data warehouse, Slack, and meeting transcripts; the campaign agent reaches Salesforce, HubSpot, and Swoogo. An agent is only as useful as the systems it can read.
- Put it on a schedule. The reporting agent runs Sunday evening so the review is ready Monday. Scheduling is what turns a tool you have to prompt into a workflow that runs itself.
- Keep a human on validation. The agent drafts the numbers and the focus areas; a person checks them and owns the call. That review step is the control that makes the rest safe to trust.
- Expand from the proof. Once one workflow runs, the team's time shifts to enablement and data quality, and the next workflow is easier to hand over.
The requirements are modest: one chosen workflow, the system connections it needs, and a named person to validate the output. That is a budget and an org decision more than a technology project, which is the part most CMOs should be funding this year.
Frequently asked questions
What is an AI-native marketing team? An AI-native marketing team hands entire workflows to agents. An agent runs a whole process end to end: it gathers data across systems, does the work, and leaves a finished result, while the marketers move to validation and data ownership. That is a step beyond an AI tool that drafts a single email and leaves the rest manual.
What did Anthropic's marketing team automate with Claude Cowork? Two workflows. A weekly metrics review that runs on a scheduled agent, which reads transcripts, checks Slack, queries the data warehouse, and drafts focus areas, cutting one to two days of work to about two hours. And campaign setup across Salesforce, HubSpot, and Swoogo, which compressed days of manual work into hours.
Which marketing workflow should you automate first? Start with a recurring, data-heavy workflow, usually weekly or monthly reporting. It repeats on a known schedule, it spends most of its time gathering data from several systems, and it produces a defined output. That profile is exactly what an agent runs well, and it gives you a clean before-and-after to measure.
What is the difference between an AI marketing assistant and a marketing agent? An assistant responds to a prompt and hands control back, speeding up one task a marketer still owns. An agent runs the workflow on its own, gathering context, taking the steps across your systems, and producing a deliverable. The assistant saves minutes on a task; the agent takes the task off the desk.
What do marketers do once agents run the workflows? They move up a level. Anthropic's team spends less time clicking through systems and more time on enablement, validation, and the data and processes the whole company relies on. The work becomes checking the agent's output, maintaining data quality, and helping more people across the business run their own programs.
What a CMO should decide this quarter
Pick one recurring marketing workflow that eats a day or more of someone's week in data-gathering, and ask what it would take for it to run itself. Anthropic's marketing operations team answered that question for weekly reporting and campaign builds, gave an agent the system connections and a schedule, and kept a person on validation. The result was days of manual work turned into hours, and a team that now spends its time on judgment and data quality. The tools to copy it are already in your stack. The decision is which workflow goes first.
We write about what it takes to run an AI-native go-to-market org, down to the workflow and staffing calls a CMO actually makes. Download the human-writing skill, the free method behind every piece here, or subscribe for the next one.
Sources
- Anthropic, "Using Claude Cowork in marketing operations to automate reporting and campaign building" (2026). claude.com
- Anthropic, "Using Claude Cowork for marketing ops: run a weekly review that preps itself" (tutorial, 2026). claude.com
- Anthropic, "How Anthropic's marketing team uses Claude Cowork" (webinar). anthropic.com
- McKinsey, "The State of AI" (2025). mckinsey.com
- Gartner, "40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, Up From Less Than 5% in 2025" (August 26, 2025). gartner.com
By Christopher Kliebenstein. We build and run AI-native workflows for operators who are shipping AI to production.