< Back to Carve Hub

Week 5: AI assistants & AI agents

US/EMEA cohort: Tuesday, Aug 5 at 9am PT | 12pm ET | 5pm UK
Pacific cohort: Tuesday, Aug 5 at 4pm PT | Wednesday, Aug 6 at 9am AEST

Overview

In this session, we explored creating and customizing AI assistants using the MUTT framework, evolving into autonomous AI agents, their platforms, future trends, and the open questions that arise around autonomy, safety, and integration.

Quick Links
📺 Session recording
🌟 Key points
💻 Session slides
📚 Further resources

Key points

  • AI assistants are customized large language models (LLMs) like ChatGPT, Copilot, and Gemini, available on paid plans (except Gemini Gems, which are available on free plans too!).

  • They improve output quality by providing consistent instructions, save time by removing repeated prompts, and can be shared with team members for consistency. They’re trained on specific contexts, instructions, and documentation.

  • The MUTT Framework for building AI assistants:

    • Mission: Clearly define objectives, context, and tone. Think your COIF prompt and more!

    • Uploads: Provide examples, documentation, policies, or website content for training.

    • Test: Stress test with prompts; identify areas to improve.

    • Tinker: Refine instructions and iterate until quality meets requirements. Anytime your AI assistant annoys you, that’s a great time to jump in and edit your instructions to make sure the mistake doesn’t happen again!

  • What are AI agents? Unlike assistants and LLMs generally, agents act autonomously, learn over time, and can operate solo or in teams. They require a clear task, a LLM, and tool integrations.

  • Key agentic AI concepts:

    • Multi-agent systems: Specialized agents managed by a central “manager” agent.

    • Vertical agents: Industry-specific with deeper expertise.

    • Human-in-the-loop: Requires human oversight.

  • ChatGPT Agent 101: uses a virtual computer to perform web-based tasks, can access your online accounts (email, calendar, LinkedIn, CRM, etc) when granted permissions. Can create PowerPoints and Excel files for first time

    • Major current issues: slow, buggy, stopping/starting to log in or greenlight steps feels like a distraction

    • However the potential is HUGE and really exciting!

  • Questions and futurespective: expect advanced researcher agents, AI-led organizations, and radical changes to the workforce. Key questions involve agent autonomy, safety, data privacy, and integration with current systems.

Session slides