Guide · beginner
How to give the AI agent its first goal
Give Montr BI's AI agent a plain-language goal, answer its questions, watch it work and approve or reject its actions. A safe first run in about 8 minutes.
You know the outcome you want. You do not want to click through ten screens. The AI agent turns a goal into a plan and does the work, and asks you before anything risky. This guide runs a safe first goal.
What you need
- A Montr BI account where the agent is included in your plan.
- The tools the goal needs, connected. The agent can only act on what is connected.
- A small, low-risk first goal.
Step 1: pick a good first goal
Good first goals are specific and safe. They read data and make drafts.
- “Summarise my open deals and tell me which have gone quiet this week.”
- “Draft follow-up messages for the five oldest quiet deals. Don’t send them.”
- “Look at my last 30 days of numbers and tell me what moved.”
Avoid a first goal that sends messages, makes calls or spends money.
Step 2: choose where to ask
You have two places.
- AI box. For quick requests. Open it with Ask AI in the header, with Cmd/Ctrl+J, or from the ask bar on Home.
- Agent workspace. The full page for longer jobs. Work you give it here is tracked as missions.
For a first goal, use the Agent workspace. You get a full timeline.
Step 3: describe the outcome
Type your goal in plain words, as you would brief a new assistant. Include:
- What you want.
- Which records or channels are involved.
- What to leave alone (“don’t send anything”).
Example: “Find deals with no activity for 14 days. List them with the last contact date and suggest one next step for each. Don’t contact anyone.”
Step 4: answer its questions
The agent may ask follow-up questions if the goal is unclear: which pipeline, which time window, which brand. Answer briefly.
Step 5: set how much freedom it has
You can set an autonomy level per mission. For your first goal, use Watch or Approval-first, so that every action asks you.
Some actions always need your approval, whatever level you pick: WhatsApp sends, voice calls, campaign sends, destructive CRM actions and spending on image, video or audio generation.
Step 6: watch it work
Open Live activity to see what the agent is doing in real time. Each mission has a timeline of its steps, and a control room view shows what is running, what waits for you, what it cost and what failed.
Every mission has hard limits on tool calls, time and credits. If you want to stop, press Stop, and the mission halts at its next tool call.
Step 7: approve or reject in Tool approvals
When the agent wants to do something that needs your yes, it appears in Tool approvals. Each card shows the full action: the message text, the recipient, the draft.
For each one you can:
- Approve. It goes ahead.
- Reject. It does not.
- Delegate to a teammate.
Read before you click. Approvals expire if you leave them too long.
You can also see a count of waiting items on Home, in the Needs you section.
Step 8: read the report
When the mission ends, it writes a report. Find it in the Agent workspace folder inside Documents, under Reports. The agent also keeps notes on your business in an Agent Memory document you can read and edit.
Check the work against your CRM. Did it find the right deals? Were the drafts on-brand?
Step 9: widen its freedom slowly
If the first run went well:
- Try a goal that creates drafts you approve.
- Then allow single low-risk actions, such as tagging contacts, to run without asking.
- Keep sends and spending on approval.
What the agent will not do alone
- Switch on an automation it built. You activate it.
- Launch an ad. Any campaign it proposes is created paused.
- Run past its limits.
Text from outside, such as customer messages and web pages, is treated as data, not instructions.
Common mistakes
- A vague goal. “Grow my business” gives vague results. Be specific.
- Not connecting tools. It cannot act on what it cannot reach.
- Approving without reading. The card shows the real action. Use it.
- Starting with sends. Begin with reads and drafts.
Next steps
Read how approvals work in Montr BI and why your AI should ask before it acts. Learn more about the AI agent.