Procore's Agent Builder helps users create AI Agents to automate routine tasks and gain predictive insights from project data. You can easily tailor the Agent to meet your specific project requirements and business needs by selecting the actions it should take and the data it should access. After creating the Agent, you can run it directly from Assist, Procore's in-app AI service.
on the Create Agent button and then choose Import Procore Agent. Select an .md file from your device. This is a Markdown text file with a short block of setup information at the top, called “front matter”.When members of your project view the available Agents in Procore Assist, they can see each Agent's title, description, and author.
Choose what documents and actions power this Agent.
Attach documents from the Procore Documents tool that you want the Agent to reference when executing its tasks.
Note: Documents must be .pdf and 50 pages or less.
The Agent can perform certain actions for you within Procore like creating an observation, sending a task to someone, or accessing data.
Write the instructions the same way you would when requesting something from a coworker. Tell the Agent what information or task you'd like to achieve and how to use any attached documents. Include as much information and detail as you can and use complete sentences. When someone runs your Agent, these are the instructions the Agent will follow.
Review the following best practices for writing Agent instructions:
1. Start with the most important points. Begin by stating your main goal or question, and then give any critical constraints (example 1, example 2, etc.).
2. Keep it simple
3. List out specific steps or details. If the task requires multiple steps or details, grouping them in a short list can help. For example:
Step 1 - Detail 1 - Detail 2 - Detail 3 Step 2 - Detail 1 - Detail 2 - Detail 3
4. Use examples (if helpful). Quick examples act like mini templates for your instructions (e.g., "Here's a sample answer style that I like...").
5. Mention format requirements. If you need the output in a specific format, mention that explicitly. For example, say "Provide your answer as bullet points."
6. Be brief, but complete. Too much text can be confusing. But if it's too short, the AI may not know exactly what you mean. Aim for a sweet spot--just enough detail to cover your request without burying it in extra text.
7. Add notes for other users. To include comments (information for yourself or teammates, not for the Agent to read or execute), begin the line or paragraph with two forward slashes. Example: // This Agent consolidates morning activities into one efficient interactive briefing for daily field operations.