Profound’s October 16 launch post for its rebuilt Agents has one line that tells you how to read the whole product. The new Agents “reason through judgement calls using your brand and personal context, so that the human-in-the-loop is a choice, not a blocker.”
That is an honest pitch, and for a lot of the work AEO teams do every week it is the right one. The weekly visibility report, the competitor check, the note when an answer engine starts repeating something false about your pricing: the mechanical part is easy, and the judgment in the middle is what keeps landing on a person. Profound wants an agent to take that judgment too.
The catch is in who gets to make the choice. Reading the documentation, almost every control that decides what an Agent does on its own sits with an individual: the person who set the trigger, the person who clicked Always allow, the person who last edited the instructions. Very little of it sits with the team. So before you hand Profound your WordPress login, decide who owns each of those choices.
A note on evidence. This is a primary-source review. I have not used Profound Agents. It is built from Profound’s launch post, its October 16 changelog, about fifteen documentation pages and its pricing page, all read on October 20. The only outside coverage I found was a short AEO Wire item that restates the announcement, so treat the vendor’s framing as the vendor’s.
What it is
Profound sells AI search visibility: where your brand shows up in ChatGPT, Gemini and Google’s AI answers, which sources get cited, and what the engines get wrong about you. Its assistant, AI Marketer, or Aim, sits on top of that data.
An Agent is a recurring job you hand to Aim. You describe it in a sentence or two, Aim asks a few questions, then writes the instructions and picks the tools. The getting started guide uses a weekly check on whether visibility dropped, saved as a short summary. Each Agent has one page with a chat on the left, where every run happens in view, and a settings panel on the right.
The tool list is where this stops being a reporting feature. Agents can write to WordPress, Webflow, Contentful and Sanity, to HubSpot, Klaviyo and Customer.io, to Gmail, Slack, Notion and Google Drive, and to any service with a public MCP server you add as a custom connector. An AEO agent can now touch your website and your customer messaging stack.
What changed on October 16
Profound already had Agents. Those were node-based automations you built step by step, and the launch post admits that “when anything unexpected happened, the Agent failed.” They are now called Workflows and keep running unchanged. The new Agents reason through each run instead of following nodes, and you can migrate a Workflow into one with a click while keeping the original.
The bigger change is triggers. An Agent can now start itself on a schedule, on a Slack or Teams event, when a web Monitor spots something new, such as a competitor post, or when a Profound metric crosses a line, including a rise in inaccurate claims from its FactCheck feature. The changelog says the new Agents are rolling out to all Enterprise and self-serve customers. The FAQ calls them a public beta, and during the beta they cost no credits.
What works
The build is a brief, not a diagram. Aim asks one clarifying question at a time, and anyone who has inherited a node graph from a departed contractor will see the appeal of instructions written in plain language.
The work is visible. Every run is a chat with its reasoning, tool calls and results, saved under Run history. You can ask follow-up questions about a run in the same place you would ask for a change.
The triggers point at the right things. “Inaccurate claim count went up” is a far better reason to wake a person, or an agent, than “it is Monday.” So is a competitor shipping a comparison page that targets your category.
Profound kept the deterministic option. If a step must run the same way every time, like a calculation or a fixed publishing path, an Agent can run a Workflow as one of its steps. The comparison page is candid about the trade: Agent output “can differ between runs,” while a Workflow returns the same structure every time.
There are real guardrails to build with. Agents have version history with restore. The roles page includes an Agent Operator role that can run existing Agents but cannot create them from scratch or edit them. Enterprise admins can cap credits per member. And the FAQ says Agent data is contractually barred from training third-party models.
What breaks or is fenced
Triggers belong to a person. This is the design choice that matters most. Per the triggers guide, a trigger belongs to whoever set it, “only that person sees the triggered run and its outputs,” only that person can flip the switch that pauses it, and triggers are not saved in version history. Share an Agent with five people and each can run it, but if one of them set the Monday 9 a.m. trigger that posts to the brand channel, only that person can see those runs or pause them. The docs do not say what happens to that trigger when its owner leaves or goes on holiday. Ask.
Edits go live on the next run. “There’s no separate publish step,” the getting started guide says, and the FAQ adds that “there are no separate test runs.” Anyone with Can edit access changes the instructions, and the next triggered run uses the new version. Version history lets you restore afterward. That is a fix after the fact, not a review before it.
Approval is a setting, not a step. By default an Agent asks before it changes anything in a connected service, and the approval card shows up in the run’s chat. Permissions are granted per chat, so a triggered run asks again, and the FAQ’s advice for stopping the repeat prompts is to pick Always allow, set the action to Allow in the tool’s settings, or set the whole Agent to approve automatically. The per-tool setting “applies to all your Agents.” There is also a switch that lets an Agent edit its own instructions. And the FAQ says plainly that Agents can publish to your CMS “without a manual approval step,” noting that most teams start with review and automate once confident.
The formal gate Profound documents, a Human Review node that pauses until someone approves, rejects or edits the text, belongs to Workflows. An Agent gets it only by running a Workflow that contains one. And if that node’s reviewer is set to “any member,” the first person to respond decides.
The inputs come from outside. A Monitor reads the open web. A Slack trigger starts a run from a channel message, and Profound’s Slack guide notes that anyone who can see a thread with Profound can reply, and Aim answers with the access of whoever started the thread. My reading, not a reported incident: text you do not control should not flow straight into an agent that holds publish rights to your site or send rights to your email tool without a person in between.
Context updates itself. Context Manager, which supplies brand context to Agents on Enterprise and Agency plans, maintains brand knowledge records on its own and favors newer information from trusted sources, like a customer call or an approved Slack conversation. Your own edits win. But the brand facts an Agent writes from can shift between Monday and Tuesday without anyone opening the page.
Who it is for right now
Teams already paying for Profound who run AEO as a weekly content program, with a named owner for the CMS and a habit of review. Agencies, too: the launch post’s example of one run per client for monthly reporting is a clean fit, and Profound’s self-serve Agency Growth plan comes with 400 credits per month per client workspace.
It is a weak fit for teams that want publishing to be deterministic, for anyone without someone who will read run histories, and for teams that would connect Klaviyo or HubSpot because the tool list allows it rather than because a job needs it.
Pricing and limits
Agents are free to run during the public beta. After it, the FAQ says, “an Agent run uses more credits than the same task run as a Workflow,” and “an hourly monitor can start up to 24 runs a day,” each a full run. Profound publishes no price per credit. Its pricing page lists a free Trial with a one-time credit allotment and custom Enterprise packages, with overage billing or a hard pause depending on how the account is set up.
Two details are worth checking with your account team. The pricing FAQ promises an estimated credit cost before an Agent runs, while the credits documentation describes estimates only for Workflows and Sheets. And per the teams guide, individual credit limits stop applying once a member spends from a team wallet.
Other limits: Agents cannot be run through the Profound API or MCP yet, cannot run in bulk from Sheets, and cannot call other Agents. A web Monitor’s query and frequency cannot be edited after you save it; you delete it and add a new one.
What to do this week
Profound Agents are worth a pilot if you already pay for Profound. Use the free beta to learn the shape of the work, and set the rules before billing starts.
- Map roles first. Give most people the Agent Operator role and keep Member, which can create and edit Agents, for the few who own the program.
- Route every publish through a Workflow. Anything that writes to WordPress, Webflow, Contentful, Sanity, HubSpot, Klaviyo or Customer.io runs as a Workflow step with a Human Review node, with the reviewer set to the person who runs it rather than any member.
- Keep a trigger register. For each trigger, list its owner, what fires it, what it can touch and who covers when that person is out. Add trigger hand-off to your offboarding checklist.
- Set approval defaults on purpose. Leave publishing and sending actions on Ask or Block in Tools and permissions, and leave self-updates off unless you have a reason.
- Treat Monitor and Slack triggers as untrusted input. Agents fired by them should draft into a Document, not publish.
- Run a credit dry run. Count every trigger’s likely runs per month now, and ask Profound for the post-beta rate and whether Agents show an estimate before running.
- Lock the house facts. Correct the Context Manager records on pricing, claims and product names by hand, because your edits outrank everything else.
Profound has built a good place to hand off the weekly AEO grind. Make sure the hand-off goes to the team, and not just to whoever clicked first.
