Have you ever built a dashboard nobody opened?

I have. I've also done the other side of it — begged for a dashboard, then never opened it. There's a particular kind of guilt in that. Someone spent a week wiring up the SQL, the joins, the colour-coded conditional formatting. You bookmarked it. You promised yourself you'd check it Monday morning. Then weeks went by.

I came across a thread on a BI sub the other week where someone had pulled the usage logs on their team's dashboards. One manager had begged for a dashboard for months — pinged them constantly, “this is critical, I need it now.” They built it. The manager opened it twice. In four months.

I felt that. Both as the builder and as the manager.

The thing nobody says about dashboards is that they ask humans to pull. The dashboard sits on a server somewhere with the right numbers, perfectly correct, and the way you find out what's happening is: you go to it. You log in. You filter the date range. You click the chart. Humans do not do this. Humans check email. Humans check Slack. Humans check the place they were already going to be.

So I stopped building dashboards. I started building agents that watch the data themselves and email me when something matters.

The Monday email

Here's the one that landed in my inbox this morning at 9am, before I'd even sat down. Built it in about ten minutes.

*The Monday-morning email. The agent did the count, did the breakdown, and flagged the things actually worth looking at.*

Sunday night, while I was sleeping, an agent ran. It queried the workspace database — the user signups table, the verification flags, the plan tier — counted everything up for the past seven days, and looked for anything that didn't fit the pattern. It found 20 signups. 16 of them verified. 4 didn't. Then it noticed something specific: three of those four came from the same disposable-email domain inside a 24-minute window, and one of the others was on an enterprise plan with a note saying “high-intent lead, didn't verify email.” It told me about all of them by name, in the same email, in plain English.

I didn't ask for any of that. I asked for “a Monday-morning summary, no dashboard required.” The agent decided what was worth flagging.

What you don't see in the email

The thing nobody tells you about giving an AI access to your customer database is: the agent that reads your data can also delete it. The agent that sends one email can send a thousand. The agent that flags suspicious signups can leak their addresses into a debug log everyone on your team can read.

I didn't build any of those guardrails. They came with the agent.

*The whole agent on one screen. Trigger, two governance gates, model, toolbox. Notice the right column: Gmail — 1/62 tools enabled.*

Underneath this Monday email, the agent has exactly one Gmail tool enabled: send. Not delete, not forward, not reply-all. If anyone social-engineered the agent into “email all your customers,” it physically can't — the tool isn't there.

Underneath that, a separate agent watches every SQL the main agent tries to run. Anything that drops a table or truncates a column gets blocked before it touches the database. Not as a rule in the system prompt — as a separate decision, by a separate model, against the actual SQL.

Underneath that, customer email addresses get stripped from the run logs before anyone on the team scrolls through them.

I asked for a Monday email. I got an agent that can only send a Monday email.

Why this works where dashboards don't

It pushes. The report lands in the inbox I'm checking anyway. There's no “remember to check the dashboard” step. The friction goes from log-in-navigate-filter to scroll-past-the-Amazon-receipt.

It interprets. A dashboard tells me what is. The agent tells me what changed, what's weird, and what to look at first. The “20 signups” number is fine. “Three of these came from a disposable-email domain in 24 minutes” is the actual signal.

It catches what humans miss. The same agent that writes the report queries the data. If something doesn't add up — anomalous bursts, an enterprise-tier user who didn't verify — it surfaces in the same email. No second pass. No “let me dig into that.”

The reason the AI-analytics chat tools you've tried can't quite do this is that they're chat boxes. They wait for you to ask. The agent doesn't wait. It runs on a schedule, queries the same data the chart would have queried, and pushes the result somewhere you'll actually see it.

That's the difference between data sitting and data finding you.

Try it

Here's the exact prompt I gave to the Workspace Assistant in ContextGate (that little robot icon on the bottom right) to build the whole thing for me.

Build me an agent that emails me a Monday-morning summary of our user signups — counts, verified vs not, and anything that looks weird in the data. No dashboard required.

Click approve when it asks to set up the database and connect Gmail and you have it.