Guidelines for Creating Custom Agents (15636)-Guidelines for Creating Custom AI Agents Recording

Video Transcription

Video Summary

The webinar introduced custom AI agents for engineering firms, led by Jordan Meadows of C1S Group, and focused on practical ways to improve efficiency, consistency, and access to company knowledge. After opening remarks from ACEC staff and coalition leaders, Jordan explained that custom agents are specialized versions of large language models configured for a specific role, workflow, audience, and output style. He emphasized that AI should augment, not replace, people, and repeatedly stressed the importance of human review.<br /><br />The presentation reviewed how large language models work, their strengths, and their weaknesses. They are especially useful for summarization, brainstorming, drafting, and pattern recognition, but they do not inherently know truth and can produce inconsistent or incorrect results. To improve reliability, Jordan recommended strong prompting, clear guardrails, limited knowledge sources, and specific instructions about what to do when information is missing.<br /><br />He then shared best practices for creating agents: define one agent for one task, use clear names and roles, specify tone and audience, connect relevant reference materials, ask for citations, and test carefully before company-wide rollout. Ethical and governance concerns were also highlighted, including privacy, copyright, access control, and keeping sensitive data out of broadly shared tools.<br /><br />Jordan’s live demo built a meeting-minutes agent in Microsoft 365 Copilot. The agent extracted decisions, action items, due dates, and unresolved questions from transcripts and organized them into a structured, easy-to-scan format. He closed by encouraging attendees to start small, solve real workflow pain points, and iterate over time.

Keywords

custom AI agents

engineering firms

large language models

human review

workflow automation

prompt engineering

AI governance

Microsoft 365 Copilot

meeting minutes agent

knowledge management

efficiency improvement