Weekly Performance Brief
Slug: ai-weekly-performance-brief · Category: Operations & Insights · Provisioning: Instant
Emails the business owner a short brief on what their agents did last week, how that compares with the week before, and one thing to do about it. The failure it prevents is the quiet one: agents running for months while nobody looks at whether they are working.
Required credentials
| Provider | Used for |
|---|---|
anthropic | Turns the already-computed figures into a readable brief and one recommended action |
No third-party data source is involved, and no mail account is needed: every number comes from AgentLane’s own records for that client, and the brief is sent from AgentLane’s own mail account. That makes this the smallest credential set in the catalog.
Who it comes from: the sender address is
contact@agentlane.agency, with the client’s business name as the display name. The address is on AgentLane’s domain, not your agency’s, so a white-labelled client can see it in the email header.
Which AI actually runs this: AgentLane’s own managed Google Gemini models on Vertex AI. No Anthropic or OpenAI key is read for inference. The
anthropicrow above is a legacy credential requirement kept so existing rows and portal chips keep working — it is never checked before provisioning. See AI Models & Credits.
Industry fit
Every vertical, and honestly so — this agent reports on execution, conversation and follow-up records that exist identically for every client regardless of what the business does.
How to deploy
It provisions instantly, without staff review — request it as in Deploy your first agent. There is nothing to call: AgentLane triggers it itself, at 8am on Monday in the client’s own timezone, and sends it to the client’s portal user. A client with no portal user is skipped rather than producing a brief with no addressee.
Flow
Weekly trigger ──► Look up client config ──► Fetch this week's metrics ──► Fetch last week's metrics ──► Compute deltas (code) ──► Write brief (Gemini) ──► Email brief (AgentLane mail)Both metric fetches hit AgentLane’s own callback endpoint, which returns counts it computed in the database — the flow never runs a query of its own. The deltas, percentages and success rate are then all arithmetic in a code step, and the model is handed the finished numbers with an explicit instruction never to derive another. A brief that quotes an invented figure is worse than no brief at all.
If nothing ran at all that week, the brief says so plainly and treats it as the finding.