Working with OpenTag

Asking for work

How to phrase a job so OpenTag finishes it the way you meant, and what it does when your ask is ambiguous.

You don't need prompt engineering. Ask the way you'd ask a competent new hire who doesn't yet know your shorthand.

That comparison does most of the work in this page. A new hire doesn't need you to specify a format, but they do need to know what "done" looks like. They'll ask if something's ambiguous. And they'll be much better in month two, because by then they know what you meant last time.

Say what done looks like

The single biggest difference between a good result and a disappointing one is whether OpenTag knew what you were going to do with it.

Look at our churn.

versus

Look at our churn last quarter. I'm presenting to the board on Thursday and I need to know whether it's a pricing problem or a product problem.

Same data, completely different work. The second one tells OpenTag what question it's actually answering, and it'll come back with an argument rather than a chart.

Name the tool when it matters

OpenTag will pick a source on its own, and usually picks well. Name one when you care:

  • "in Linear, not GitHub"
  • "from Stripe, not the spreadsheet"
  • "the HubSpot deal, not the thread"

If the tool isn't connected, OpenTag tells you and hands you the link rather than quietly working around it with worse data.

Ask for the work, not the steps

OpenTag plans its own approach. Handing it a procedure usually makes the result worse, because you've constrained it to your guess at the method.

Export the CSV, then pivot by cohort, then chart it.

Show me retention by signup cohort for this year.

The second gets you the analysis. The first gets you a chart, and only that.

The exception is when the method is the point: a compliance process, a format someone else requires, a step your team always does. Say so, and OpenTag will follow it. If you find yourself saying it repeatedly, that's a skill waiting to be written.

Say what you don't want touched

Especially early on. "Don't message anyone", "read-only", "draft it but don't send" all work, and OpenTag honours them.

You have this guarantee anyway, because approvals stop irreversible actions regardless of what you asked for. But saying it up front means OpenTag plans a job that doesn't need the approval in the first place.

When OpenTag asks you something

If your ask is ambiguous in a way that changes the work, OpenTag asks before starting rather than guessing.

You@OpenTag can you follow up with the people from yesterday's calls?

OpenTag

Two readings, and they're different jobs:

  • Draft follow-up emails for you to review and send, or
  • Log the calls in HubSpot and file the action items in Linear

Which: or both?

Answer in the thread. It doesn't re-ask.

If it's asking about something you don't care about, say so ("your call", "whatever's faster") and it'll decide and move.

Follow-ups are cheap

You don't have to get the ask perfect. OpenTag works in a thread, and "no, by team not by person" or "shorter, and drop the caveats" is a normal way to steer it. A follow-up costs a fraction of the original run.

Corrections stick

Tell OpenTag a fact about how your company works: "revenue always means net of refunds", "Priya owns anything billing". And it keeps it. Next month you won't be repeating yourself. What OpenTag remembers covers what it keeps and what it doesn't.