intelligencesupport

AI agents

AI agents read your data on a schedule and report what changed. This covers standard and custom agents, scopes, brand choices, findings, reports, and costs.

For
Everyone, especially admins
Updated

AI agents are background workers that read your data on a schedule, compare it with what they saw last time, and tell you what changed. Any member can keep personal agents. Admins run the shared ones for the whole organization.

Opening the Agents area

Click Agents in the header, next to Playground and Skills. On a small screen the same link sits in the menu drawer. The area has four tabs:

  • Personal: your own agents.
  • Organization: the shared agents, for admins and brand managers.
  • Agents Command Room: what is running now, the agents switched on with their last and next run, and the previous runs, scheduled or started by hand.
  • Catalog: every standard agent, open to everyone to read and add to their own agents, and the tailor-made ones.
The Personal tab of the Agents area, with the four tabs on top and one card per agent
The Personal tab of the Agents area, with the four tabs on top and one card per agent

What an agent does

An agent works in passes. On each pass it reads one or more data sources for a brand, such as tracked Google rankings, live Google Ads campaigns, or the sales pipeline. It then compares that data with a snapshot of its previous pass. If something moved, a model reads the data and writes up what changed and what to do next.

If nothing moved, a watcher skips the model call and the pass costs nothing. Same when a brand has no data for the agent yet: the pass ends before any model call. The one exception is a custom agent that reads a connected system, since the model does that reading: the pass shows what the reading cost.

Agents hand back one of two things. Watchers write findings: short alerts with a title, an explanation and a suggested action, up to 10 per run. Report agents write a report, a longer document such as the weekly operating review, and their top signals land as findings too.

Each finding carries a severity. critical means money or traffic is being lost right now, warning flags a negative trend, and info covers positive or neutral signals worth knowing.

Standard agents and custom agents

Standard agents are built into the app, one per data source. There are thirteen so far: seven watchers (search rankings, AI visibility, Google Ads campaigns, social publishing, LinkedIn publishing, LinkedIn Ads, and the market monitor), two report agents (the weekly operating review and the executive report), the Web editor, which writes and publishes the pages of your websites, the Verification codes agent, which files away the code emails of the mailboxes you connected, the Library quality agent, which lists the pictures and clips too small to publish well, and the Podcast and YouTube listening agent, which reports every time your brand or a competitor is named in the shows you follow. The watchers run about once a day or every two days, the Web editor and the Verification codes agent daily, the operating review weekly and the executive report monthly, and you can change each schedule. More agents appear in the catalog as new data sources are added. The full list is in the agent catalog.

Note: Standard agents are opt-in. Nothing runs until someone adds the agent: a member adds it to their personal agents from the catalog, or an admin adds it to the organization from the Organization tab.

Custom agents are the ones you build yourself. You pick the data they read from the app’s modules, your Assets Library, and the systems your organization connected, then you write their instructions. From then on they behave like standard agents, budgets and findings included. The how-to is in custom agents. One kind of custom agent doesn’t report at all: a List builder agent fills an email list with the people its instructions describe, and is created from the Lists page of Email.

Then there are tailor-made agents, which aren’t part of the app at all. BearingBridge builds each one on request, for one organization. They are listed in the catalog with a contact link.

What an agent may work on

Every agent has a scope: the brands it reads, the connections it works on (your mailboxes, your LinkedIn ad account, a system your organization connected), the modules and features it reads or changes, and the kind of AI model it runs, if any. Each agent comes with the scope it was built with, and the Scope tab of its form shows it whole. An administrator, or the person who owns the agent, can change any part of it at any time: switch a module off, hold a connection to reading, add a feature, or give the agent another kind of model. A run never reads or writes anything outside its scope, and you can only give an agent what you can use yourself. See the Scope tab.

Scopes: organization, brand and personal

Scopes take a minute to get used to. Every agent setting lives at a scope, and the scope decides who may change it and whose money it spends.

ScopeWhat it coversWho can change itWhose money
OrganizationEvery brand with the agent’s module on, on a scheduleAdminsThe organization’s
BrandOne brand, tuning the organization’s agentsAdmins and brand managers, for any brand of the organizationThe organization’s
PersonalYour own agents, on their schedule and in the runs you start yourselfYou, and the admins of your organizationThe agent’s owner

Scheduled work happens at the organization scope. Once an admin adds an agent there, it runs on its own clock for every brand that has the agent’s module switched on.

The brand scope is a tuning layer. Say one brand sells in a market where small ranking swings matter: it can give the rankings watcher stricter instructions for itself. A brand can also switch an agent off or pick another model. Any field left empty inherits from the organization. A brand cannot add new agents or own a custom agent.

Personal agents are yours. Switched on, they run on their own schedule on your organization’s brands, as you and billed to you, and they join the runs you start by hand, on top of the organization’s agents. They’re a good place to try an idea before it goes to everyone. Admins can also manage a member’s personal agents on their behalf.

Budgets are the one exception to inheritance. Each agent has a monthly and a weekly budget, and the tightest one along the chain wins. A brand or a person can narrow what was set above them, never widen it.

Every field of the settings form is explained in organization and personal agents.

None, one or several brands

If your organization has a single brand, you can skip this part. By default an agent works brand by brand, with one run and one set of results for each.

At the organization and personal scopes you can change that under The brands it reads. Pick One run over the brands ticked below, then tick brands:

  • Several brands: one run reads them all together and writes one result that compares them. Each finding names the brand it is about.
  • One brand: one run reads that brand only.
  • No brand ticked: the run reads only the organization’s own data: the sales pipeline, the Assets Library, and the connected systems.

A run over a set of brands belongs to no single brand, so it is filed under your organization’s No brand workspace. Its findings also show in the feed of each brand it read. Every member works on every brand of the organization, so the brands a run read never hide its findings and reports from a colleague. Read more about brands.

Findings and reports

A finding lands in the feed of the brand it is about. Acknowledge it to mark it as seen, Resolve it once it is handled, or Dismiss it if it is noise. Resolved and dismissed findings leave the open list.

The findings feed on the Market Monitor page, each finding carrying a severity, what the agent noticed and the action it suggests
The findings feed on the Market Monitor page, each finding carrying a severity, what the agent noticed and the action it suggests

Each finding also offers three shortcuts. Create task turns the suggested action into a task on the brand’s board, free of charge. Draft response has the AI write the concrete plan, brief or copy the finding calls for, and shows it under the finding. Draft post has the AI write a LinkedIn post answering the signal and saves it as a draft in your social content. The two draft actions are paid AI actions, and each draft shows its cost next to its title, after Drafted response: or Post drafted:. Using any of the three also acknowledges the finding.

The Google Ads Reporting page shows the campaigns watcher’s findings for the current brand. For the wider picture, the Watchers dashboard in Business Intelligence counts open and critical findings across every brand of your organization, and tracks how fast they get resolved (dashboards). The Agents Command Room tab of the Agents area lists each run, scheduled or started by hand, with what it found (runs and schedules). And by default, the organization’s admins get a summary email after each scheduled run that produced something.

Reports from the report agents are read on the agent’s own card, under Recent reports, and from the Agents Command Room.

Have it read to you

Each person can also have their agents’ findings, and the figures of the dashboards they pick, read to them as a short private podcast, daily or weekly. It’s switched on from the Audio briefing card at the top of Personal: see Audio briefing.

What it costs

A run that reaches the model is billed like any other AI action, and the app always shows the price, in your organization’s currencies. The Agents Command Room prints the tokens and the cost of each run, Run watchers now shows the total cost of the run next to the button, a test run shows its cost as soon as it finishes, and each agent’s settings show what it spent this month and in the last seven days. Runs that found no data or nothing changed cost nothing, unless reading a connected system was charged; the run then says so and shows that cost.

The price of a run depends on the model and on how much data the agent reads, so there is no flat rate. To get a feel for it, test the agent on a real brand first. The test prints its cost. Then give it a monthly or weekly budget, so it stops at a limit you picked.

A few agents have paid reads on top of the model call. The AI visibility watcher makes one paid data call per run for gained and lost citations. A custom agent that reads Assets Library documents pays for each download like a download from the library, and one that reads a connected system counts those reads in the run’s cost. Agent runs are also recorded in the usage log like other AI actions. See credits and costs.

Who can do what

RoleWhat they can do
ViewerOpen the Agents area, see how agents are set up and read any catalog agent in full, read-only. No test runs.
EditorAdd personal agents from the catalog or the organization, edit and duplicate them, test them, start runs, act on findings.
Brand managerEverything an editor can do, plus tune agents for any brand of the organization.
AdminAdd, edit and remove organization agents, tune any brand, manage members’ personal agents.

Nobody needs an admin to get started: every standard agent in the catalog, and every agent the organization runs, can be added to your own agents from the catalog or your Personal tab. Access to the whole area also depends on the AI Agents right in your role. See roles and rights and capabilities and modules.

Good to know

  • An agent never reads across organizations. Everything it does stays inside the organization it belongs to, test runs included.
  • Changes to an agent apply from its next run.
  • An agent that is on but never seems to run usually has nothing to work with: the brand’s module is off, or the brand has no data yet. On the Organization tab, a card warns you when no brand qualifies.
  • Findings are written for your team. They never name the model or the data provider behind them.
  • Skills set to apply to Agents are added to the runs of the watchers, the custom agents and the operating review. The executive report follows the skills set for Brand Content instead. A scheduled run uses the organization’s skills; a run you start yourself also uses your personal ones. See skills.