Introduction

  • I’m trying to work out how to use AI well
  • I’ve set up a dozen agents to work on 6-10 projects
    • [VIDEO] here’s what they look like doing it
  • everything from a stock market investor
    • [TAB] show constellation
  • to a nuclear reactor designer
    • [TAB] show reactor design
  • this is what I’ve learned

Findings

  • Give them context
  • Give them a job to do
  • Measure how well they do it
  • Repeat, with lots of them working in parallel
Falling back to PNG as your browser does not support SVG.
The loop: context, a job to do, a measure of how well it went — then round again, with many agents at once.

Single web-based chats

  • learning stuff
  • abstract, disconnected

AI where you work

  • connected
  • seeing what I’m seeing
  • access to what I’m working on

Action-based chats

  • uploading files
  • modifying and downloading files

Folder-based local chats

  • modifying local copies
  • version control
  • refer to old screencasts
    • using git
    • publishing markdown

Other environments

  • summarising email, drafting responses

Considerations

  • trust and access
  • security
  • data issues, storage, processing, GDPR

Help

Tags: ,

Updated:

Leave a comment