I've jumped feet first into the 'vibe' coding and trying a solution I believe will be impactful in my vertical, but am learning all the lessons the hard way - hallucinations, over-eager agents ruining established code, establishment of agent.md, skills.md, token limits reached, etc., etc. I'm glad to support this group and look forward to learning best practices and being more autonomous without sharing all of my data across all the major platforms. I also want o reduce my 'subscription fatigue'. It's ridiculous what I spend in all of the various subscriptions......
Wow I love your approach to AI. It’s got me thinking about it in a way I wasn’t before. Definitely going to Build a shield. Please let me know when your making Available the architecture that changes memory into computer language so we’re no longer enslaved to tokens. There’s so much I wanna build, but I’m limited.
Memory was the single biggest unlock for my agent. Without it, every session was groundhog day - same mistakes, same questions, same context rebuilding.
I ended up with a three-tier system: short-term memory (2-3 days, full details), weekly rollover (compressed summaries), and permanent index (people, projects, accounts). The agent literally gets better at its job over time because it remembers what worked and what didn't.
The bankruptcy angle is real too. Running everything on the strongest model burned through my limits in 3 days. Different memory tiers need different model tiers — you don't need Opus to recall a file path.
I've jumped feet first into the 'vibe' coding and trying a solution I believe will be impactful in my vertical, but am learning all the lessons the hard way - hallucinations, over-eager agents ruining established code, establishment of agent.md, skills.md, token limits reached, etc., etc. I'm glad to support this group and look forward to learning best practices and being more autonomous without sharing all of my data across all the major platforms. I also want o reduce my 'subscription fatigue'. It's ridiculous what I spend in all of the various subscriptions......
Wow I love your approach to AI. It’s got me thinking about it in a way I wasn’t before. Definitely going to Build a shield. Please let me know when your making Available the architecture that changes memory into computer language so we’re no longer enslaved to tokens. There’s so much I wanna build, but I’m limited.
Memory was the single biggest unlock for my agent. Without it, every session was groundhog day - same mistakes, same questions, same context rebuilding.
I ended up with a three-tier system: short-term memory (2-3 days, full details), weekly rollover (compressed summaries), and permanent index (people, projects, accounts). The agent literally gets better at its job over time because it remembers what worked and what didn't.
The bankruptcy angle is real too. Running everything on the strongest model burned through my limits in 3 days. Different memory tiers need different model tiers — you don't need Opus to recall a file path.
https://thoughts.jock.pl/p/my-ai-agent-works-night-shifts-builds