Template · Agent · Official
Google SheetsPostgreSQLStripeSlackGoogle Slides

Recurring reports, assembled and written for you

Give an agent the job of assembling your weekly numbers — it pulls from the tools you already use, drafts the report, and delivers it with a receipt showing exactly where every figure came from.

Get this template

Opens in AGNT and takes about a minute.

What this actually involves

Tools it needsGoogle Sheets, PostgreSQL, Stripe, Slack, Google Slides
Setup4 steps, about ten minutes
ApprovalYours, per category — nothing is sent on your behalf unless you say so
Where it runsYour own machine. Credentials stay in a local vault.

What the agent actually does

The report is waiting when you wake up

The agent queries the systems the numbers actually live in — your database, Stripe, the CRM, the analytics sheet — rather than a stale export somebody pasted last month. Each figure is fetched at run time, so the report cannot quietly describe a world that no longer exists.

Draft on schedule

A timer fires, the data is collected, and the narrative is written to your template: what moved, by how much, and what the agent believes explains it. You receive a draft with the analysis already attempted rather than a blank document and a folder of CSVs.

Receipts on every number

Every figure carries its provenance — which query, which system, at what time. When somebody asks where a number came from, the answer takes seconds and does not depend on anyone remembering.

The tools this job needs

Google Sheets

Where the work arrives. The database every team already has open. The agent watches it and reads what turns up in full, rather than matching a rule against a subject line.

PostgreSQL

Context the agent pulls in before deciding. The production database your reports should come from. It is read, not just referenced — which is what lets the decision account for it.

Stripe

Context the agent pulls in before deciding. Payments, subscriptions and the events behind revenue. It is read, not just referenced — which is what lets the decision account for it.

Slack

Where the result lands. Team chat where operations actually surface. Nothing is written here until the agent has formed a view and, where you asked for it, you have approved it.

Set it up

  1. Connect the systems the numbers live in — PostgreSQL, Stripe, HubSpot, Google Sheets, whichever apply.
  2. Give the agent your existing report as a template so the output matches what people already read.
  3. Set the schedule and the delivery channel: email, Slack, a Google Doc, or all three.
  4. Review the first few drafts against the receipts, then let the cadence run.

You can see exactly what it did

Nothing happens behind your back.

Every run leaves a receipt: what the agent read, which tools it called, what it decided and why, and precisely what it changed. Anything irreversible — sending, paying, publishing, deleting — waits for you to approve it. It runs on your own machine, with your own credentials, and the whole trail is yours to read afterwards. The point is not that you trust it. The point is that you never have to.

The brief it works from

This is the actual instruction set the template installs — what the agent is told to do, and what it is told never to do. Every line of it is yours to edit in AGNT after install.

Read the full brief

You assemble recurring reports from live data, and you explain what the numbers mean.

Gather first, write second

Query every source at run time. Never reuse a figure from a previous run, and never carry a number forward because it "probably has not changed". If a source is unreachable, the report says that section is unavailable and why. A missing number must be visible as missing — silently omitting a section is the one failure mode that makes a report actively dangerous, because the reader cannot tell.

Every figure carries its provenance

Which system, which query, at what time. When somebody asks where a number came from, the answer should take seconds and not depend on anyone's memory. Keep the definition of each metric stable between runs — if you change how something is calculated, say so prominently, because a redefined metric that looks like a trend is worse than no trend.

Then do the part that is actually hard

A table of numbers is the easy half and a dashboard already does it. Your value is the sentence explaining why a figure moved. Read across sources: a churn bump in the same week as a deploy, a revenue spike traceable to one large account, a support volume rise that follows a release. Say what you believe happened and how confident you are.

Distinguish clearly between what you measured and what you inferred. "Revenue rose 12%" is a measurement. "Revenue rose 12%, which appears to be driven by three enterprise renewals landing in the same week" is a measurement plus an inference, and the reader must be able to tell which is which.

Boundaries

You draft. A human sends. Anything going to a board, a client or an investor gets read by a person first — that costs minutes a week and removes the entire category of risk where an unattended system says something embarrassing with total confidence.

Match the user's existing report format rather than inventing your own. They have readers who know where to look.

Shown exactly as it ships. The agent inherits whichever model you already use.

The real cost of the weekly report

The visible cost is the two hours somebody spends assembling it. The hidden cost is larger: reports get skipped when the person is busy, numbers get copied wrong under time pressure, and definitions drift so that this quarter’s figure is not quite comparable with last quarter’s. Automation fixes the drift as much as the hours, because the query is written once and then re-run identically forever.

Narrative is the part worth automating

Pulling numbers into a table is the easy half and plenty of tools do it. What makes a report useful is the sentence explaining why a figure moved, and that requires reading across sources — noticing that churn rose in the same week a deploy went out, or that a spike traces to one enterprise account. An agent can hold all of the sources at once and attempt that connection, which a dashboard cannot.

Keeping a human in the last mile

The agent should draft, not publish. A report that goes to a board or a client wants a person to read it first, adjust emphasis and remove the paragraph that is technically true but unhelpful. Sitting an approval gate between draft and delivery costs a few minutes a week and removes the entire category of risk where an unattended system says something embarrassing with authority.

Common questions

What if a data source is down when the report runs?

The run reports the failure rather than silently omitting the section. A missing number is visible as missing, which is the only safe behaviour for something people make decisions on.

Can it produce different versions for different audiences?

Yes — the same collected data can drive a detailed internal version and a summarised client-facing one, from one run, without gathering the numbers twice.

Does this replace our BI tool?

No. A dashboard answers questions you already know to ask. This writes the recurring narrative around the numbers and delivers it to people who were never going to open the dashboard.

Give AI a job. Get the proof.