Working with your agent

Your agent brings in your data, asks you questions about the process, and explains the results. glasstarn does the counting, so you get the same numbers every time.

Starting

In your AI app, type /glasstarn, or just ask: “Use glasstarn to look at this export.” You do not need to learn any commands.

A good agent session goes like this:

1. If your folder already has an export or a glasstarn project, it starts straight away: it analyses (or continues the project and says so), saves the answer, opens the results in your browser and leads with the link. Otherwise it asks what you want to find out and what data you have.
2. It tells you the main things it found, with glasstarn’s numbers.
3. It suggests two or three questions worth asking next.
4. You choose. It saves every answer to the project so the browser shows it, but it asks before pulling data from a system or changing a setting.

What your agent will do

• Quote glasstarn’s numbers exactly, and say which filter they come from.
• Label anything it cannot prove as a guess: “this looks like a staffing issue, but the data does not show staffing.”
• Ask before it pulls records from a system.
• Keep its work in the project, so you can stop and come back tomorrow.

What it will not do

• Invent numbers, timings for steps that are not in the data, or savings.
• Claim a cause from a pattern.
• Put bulk records in the chat. It works with summaries.
• Present a proposal as approved, or a forecast as measured.

Documents

You can give your agent SOPs, policies, diagrams or interview notes. It uses them as evidence and records which document and version each point came from. glasstarn does not add another approval on top of your company’s rules for your AI app. See Privacy.

Good questions to ask

Understand

• “Where does the time go between order and delivery?”
• “What do the slowest invoices have in common?”
• “Which steps get repeated most, and why might that be?”

Improve

• “Which steps could we remove or simplify before automating anything?”
• “Which steps are good candidates for an agent, and what would it need?”
• “Draft a small pilot with success measures.”

Check

• “Did every payment over $10,000 get a second approval?”
• “Compare these invoices with our AP policy.”

Share

• “Make a short summary for my manager, with the three main findings.”
• “Export the main flow as BPMN.”

Keep going

There is no project plan here. Once you have a first look, keep asking, in your own words, for whatever you need next:

• “Turn this into five slides for Monday’s team meeting.”
• “Write up three improvement ideas, with the numbers behind each.”
• “Draft an SOP for how this process should run.”
• “Compare March with April. Did anything change?”
• “Write a short email to the AP team about the invoices that wait longest.”

Your agent uses glasstarn’s numbers and its own tools to make the slides, documents and emails. If an answer is not what you wanted, say so and ask again. That is the whole method: ask, look, ask again.

Show it on the page

Ask to see something and your agent shows it on the open page instead of drawing it in the chat: “Show me the main path”, “Show me the three main variants”, “Only claims over $10,000”, “Focus on Approve payment”, “Compare South with North”, “Clear the filters”, “Take me to Conformance”. Within a few seconds the page moves there and puts on the filter, shown as a chip you can remove. It is always the same page: your agent gives you a link only the first time, or when the page is not open, and that link opens the page as it is now, never reset.

The browser view

The browser view shows the same project. You can use it without your agent, and your agent can join later. When you change something on screen, such as a filter, the agent sees it. To let the agent read the browser project, use Connect an agent at the top of the page. See Privacy.