Next next steps

[johnml1135]

How about GRTSIV?

GRT ->

• G - Getting yourself oriented with the bots
• R - setting up the repo for the bot
• T - setting up the right capabilities for the bot

This stuff is evolving, but the dist is beginning to settle. You can set things up now and get a lot of progress. Low risk of rework

SIV ->

• S - Specifications: Markdown files with elaborate plans
  • Still a lot of movement. OpenSpec and SpecKit. Devil's advocate.
• I - Implementation: Taking the markdown files and making it code
  • This is being ripped apart right now
  • Gastown, MCP agent mail, sub-agents, handoffs, pinging
• V - Verification: how do I know it is good?
  • Linters static analyzers, testing edge cases, extra AI reviews (rule of 5), Manual reviews of test cases, CI/CD

Path forward:

• 2024: autocomplete - 20% speed increase
• 2x speed increase
  • AI helps to write plans
  • Ask it to implement the code
  • Babysit the repo as it implements small chunks
  • Manual review and nudging
  • You can do this now with established tools!
• 20x speed increase (not for the faint of heart)
  • AI helps write the plans
  • Then you throw the plans over a wall into an AI chomping mess
  • Then you have AI bots do all the verification
  • All the tooling is being written by AI in a fever pitch in the past 6 months
  • Gas Town and Agentic Coding Flywheel are two current frameworks to try
  • It only works with some languages and will break all the time
  • You will have a lot of churn with constantly updated frameworks (updated at the speed and quality of AI)
  • This will likely become standard and easy(er) to use in the next 1-2 years.