Next steps #2
Description
What we are dealing with:
- Centaurs: human - machine interactions
- Impedance match: the agent can do as much as you can, not too little, not too much
- Grow to meet the challenge: If the issue you are facing is too much for you, but feasible for the machine, you need to increase your own skill to match the problem and then you can work with your copilot. Otherwise, the AI will make slop and you won't know it.
Some things become cheaper to make, but not others:
- Signals for trust and quality: suits + tie + proper English = trustworthy?
- AI will make it look fluent, but inaccurate. Good looking slop - what signals would still indicate quality?
Building a system:
- Team design: different agent writing, testing, managing + timelines, quality control, devil's advocate, simplicity and elegance / architecture (edited)
- Which agents need to be all Human? Have human oversight or review? Be completely autonomous? What about different domains? Have a person review testing and architecture but not code?
Rolling it out:
- Pioneer: Read all the blogs, break new ground, determine what is possible, make an overall plan
- Settler: Build the foundational infrastructure for people to use. Be a champion - show the team what is possible.
- Towns person: Learn and execute the system focusing on your area of expertise