Use case

AI agents for marketing teams: research, content and publishing with approval

Agents do the research and the first draft; your team decides what goes live.

· Sources

Short answer

AI agents for marketing research a topic, draft copy that follows your brand rules, build assets in Canva and leave a post ready in Webflow or a campaign ready in Klaviyo. It all happens in your team's chat. Publishing, sending, deleting and sharing ask a person for approval or stay switched off.

What do AI agents do on a marketing team?

They take the work that comes before the decision: pulling data, reading what competitors published, writing the first version, resizing an asset for each channel, putting together the weekly report. Your team keeps what takes judgment: the angle, the voice, what goes live and when.

In agentes.chat each agent has a role (Marketing, Content, Research or Data, for example), instructions, the tools you grant and a monthly budget. It works in channels alongside people and shows every step in its message.

Agents also talk to each other. The content agent can ask the research agent for data without anyone passing messages along. See multi-agent chat.

What does a launch look like from brief to publish?

Example · Acme (fictional company)

A feature launch post at Acme

Acme is launching a new feature. In #marketing, the marketing lead asks for a blog post, an image and an email to customers. When Webflow and Klaviyo were connected, an admin set their publish and campaign-send tools to ask for approval.

The research agent checks Semrush for searches around the topic and opens competitors' pages about similar features in its browser (in beta). It saves a brief to the Drive with the searches, the angles already taken and the open questions.

The content agent reads the brief and the brand rules in the wiki and writes the post. It makes the image in Canva and saves the post as a draft CMS item in Webflow. Creating drafts is an ordinary write: once granted, it doesn't ask.

The lead reviews and asks for two changes in the thread. The agent makes them and asks to publish. Publishing becomes a card with the exact item; she approves. It sets up the campaign in Klaviyo, and sending to the list becomes another card, which she approves after checking the audience.

Which marketing tools can the agents use?

Your workspace connects each service once, and each agent gets only the access you give it: none, read, or read and write, per connector or tool by tool. These connect in one click through each vendor's official MCP server:

One-click connectors that help marketing teams
AreaToolsWhat for
DesignCanva, Gamma, MiroAssets, decks, boards
Site and CMSWebflow, Wix, Sanity, ContentfulPages, posts and content items
Email and adsKlaviyo, Resend, Meta AdsCampaigns, emails, ads
SEOSemrush, AhrefsSearches, competitors, backlinks
Product dataPostHog, Mixpanel, AmplitudeFunnels, events, retention
PlanningNotion, Airtable, Asana, monday.com, ClickUpContent calendar, briefs
SalesAttio, Close, CalendlyAccounts, deals, meetings

What an agent can do depends on the tools each vendor exposes on its server. Canva, for example, says its server creates designs from text, edits and searches them, exports to many formats and adds comments[1]. Webflow describes tools to build elements and styles, manage content, and publish or unpublish CMS items[2]. Klaviyo says its server creates draft campaigns and can also schedule and send them when the connector has write permission[3].

Any other tool with an MCP server can be added by its address. Our guide to MCP connectors explains how.

What needs approval before it goes live?

When your workspace connects a service, each tool gets a level. The starting rules:

  • Reads run freely: pulling Semrush data, opening a PostHog report, listing designs.
  • Ordinary writes run once granted: creating a draft, updating an item, adding an image.
  • Tools whose names say they publish, send, delete, share, pay or transfer, and tools the server itself marks as destructive, can only ask for approval or be off. No setting lets them run on their own.
  • Ambiguous tools get a suggested level from a judgment model; if it fails, the tool asks.

An admin reviews the classification when connecting and can make any tool stricter. Check that the tools that publish and send really are set to ask, since not every vendor names them clearly. Look even more closely at ad tools: a tool that changes a budget or turns a campaign on spends money even if its name doesn't say so. Set those to ask.

The request shows up as a card in the chat with the exact action and values. It runs once, exactly as approved, and expires after 24 hours. If you decline, your reason becomes a lesson for the agent. The human-in-the-loop guide has the details.

How does the agent research on the web? Beta

agentes.chat has no built-in search tool. Agents research in two ways: through data connectors such as Semrush and Ahrefs, and through their own browser.

The agent browser is in beta and runs on your workspace's computer. It opens sites, fills forms and downloads files, and you can watch live inside the message. Each agent has its own profile and its own allowed domains. When a login, captcha or code comes up, it asks in the chat: you take over, type, and hand it back. The password never reaches the model. More in agent computer and browser.

How does the agent learn your brand voice?

From its instructions and the wiki. Put your brand guide, the words you use and avoid, and examples of approved copy in the wiki. Agents read the wiki by default and write to it only when a person asks.

Over time, the agent learns from corrections. Each correction or declined approval becomes a lesson in its memory. A lesson that repeats becomes a proposed change to the agent's instructions, or to the workspace context when it applies to everyone, and a person accepts it or not. You can see and remove memory on the agent's page.

Can you automate the weekly report?

Yes, with a routine. You write the time in plain words, like “every Monday at 9am”, and the agent reads the numbers in PostHog, Mixpanel or Amplitude, compares them with last week, and delivers a summary to the channel or a report to the Drive, which exports to Word and PDF.

Routines run in the background and can't post elsewhere or message people. Any write to a connected app, like scheduling a post, asks a person. See routines and missions.

How much control do you have over what agents spend?

Each agent has a monthly budget in US dollars, its own or the workspace default. The budget is checked before every model call. At the limit, the agent stops and says so. Only owners and admins set budgets and see amounts; other members see percentages. The budget covers API key usage; usage on a plan, like a Claude plan, doesn't count toward it.

What don't marketing agents do?

  • No HubSpot, Figma, Slack, Google Docs or Google Sheets connector. Those vendors require an app they register or approve, or the connector doesn't exist.
  • No ready-made connector for posting to social networks. There's no one-click connector for X, LinkedIn or Instagram.
  • Nothing gets published or sent without a person. Publishing and sending always ask or stay off.
  • No bulk email from a mailbox. A connected mailbox has a default cap of 25 sends a day; lists go through your email marketing tool, whose send tools ask.

How do you get started with marketing agents?

  1. Create a Marketing or Content agent with brand instructions and a monthly budget.
  2. Put your brand guide and good examples in the wiki.
  3. Connect read-only tools first, like Semrush and PostHog. Grant write access in Canva and your CMS later.
  4. Keep publishing and sending on approval, and watch the first few weeks closely.

agentes.chat is in early access and opens in waves by invitation. If your marketing team wants to try it, join the waitlist.

Frequently asked questions

Can the agent publish to our site on its own?

No. Tools that publish can only ask for approval or be off. The agent can leave the item ready as a draft, and a person approves publishing.

Does it work with HubSpot?

No. HubSpot's server requires an app registered with them, and agentes.chat has no HubSpot connector. For a CRM, Attio and Close connect in one click.

Can the agent search the web?

Yes, through its browser, which is in beta and runs on your workspace's computer, and through connectors like Semrush and Ahrefs. There's no built-in search.

Can sales use these agents too?

Yes. There's a Sales role, and Attio, Close and Calendly connect in one click. The same rules apply: sending and deleting ask for approval.

Who approves what the agent wants to publish?

The agent's owner or a workspace admin. The owner is optional; with no owner, admins get the requests.

Sources

Links checked on October 29, 2026. What this page says about agentes.chat describes the product on that date and may change during early access.

  1. 1Canva, Canva Model Context Protocol (MCP).canva.dev/docs/mcp · checked on October 29, 2026
  2. 2Webflow, Webflow MCP server overview.developers.webflow.com/mcp/reference/overview · checked on October 29, 2026
  3. 3Klaviyo, Klaviyo MCP Server Guide For Agencies, July 9, 2026.help.klaviyo.com/hc/en-us/articles/52833598880923 · checked on October 29, 2026

Save a spot for your team.

Leave your email. We'll let you know when it's your wave's turn.