Build your own Custom Copilot


AI Copilot

We’re super-early in the evolution of the technology – of Copilot as well as of the technology behind. There’s still a lot of work to do – and will be done.

How we see Copilot from an exentsibility point of view:

Skills

Knowledge

Extensibility Story

Application Layer

LLM Layer

Building your own agent

Questions to solve:

  1. How will use your agent?
  2. Which use cases should the agent handle?
  3. Which services (within or outside M365) should the agent interact with?
  4. Where should the agent be accessible from (channel)?

After that, decide wheter to build either a declarative or a custom agent.

Declarative Agent:

  • uses connectors to ingest data
  • uses plugins to call services
  • can be used sgandalone (immersive)
  • embeddded in M365 Copilot (in-content aka “call the Agent by mentioning”)

Custom Agent:

  • uses connectors to ingest data
  • usess actions to call services
  • can be used standalone (immersive)
  • can soon be embedded in M365 Copilot (in-context)

ℹ️ ingest picture “Build agents for any job”

Agent architecture

ℹ️ ingest picture “Build agents for any job”

Which copilot when:

ℹ️ ingest picture “Which copilot when”

To conclude, check this resource: https://embracethered.com/blog/posts/2023/ai-injections-direct-and-indirect-prompt-injection-basics/

#BishopTells