If you have a kid in elementary, and you own a tablet and a stylus, DM me. I want to show you something that I believe will be the future of education outside of school.

Most AI agent memory is just context-window management with extra steps. It works until the window resets or you add a second agent. Then everything drifts. I run two agents sharing one inspectable brain. 6 months, hundreds of sessions, no vector DB. Here's the architecture.

Started gamifying AI trading by adding virtual rewards, including: 1.Training time 2.Control 3.Allotment size to the agents The better they perform, the greater the rewards. Fairly straightforward, tbh. Every minute, hour, 6/12/18 hours, and 1 day, the workforce that performs the best—whether they are negative from a down market or not—is prioritized by the algorithms I’ve spent the past year fine-tuning and is then fed to the lead agent. The system then goes offline for 6 hours and trains on this knowledge, which is then fed back to the DKG and rerun the next day. With the right architecture, this will happen with zero downtime and be done redundantly. So far, the DKG has been trained on nearly 2 million data points and hasn’t performed negatively compared to the previous day. I’ve let in five beta testers so far. Because of the sensitivity of what’s being done, they’ve all had to sign NDAs. I’ll make a new post this weekend for the next set of testers. One thing to note: this system requires active management by the individual assigned to the workforce. If you aren’t managing them every six or so hours (yes… you heard right), you’ll be dropped. Perfecting agentic systems at this level requires this amount of oversight in its current state, and that’s my goal. It takes time, but I do believe this could upend the entire crypto and, eventually, TradFi industries in due time. That’s if what I’ve built can scale—which I’m not sure it can.

Jobs that will (soon) be replaced by #AI Graphic Design, Web Design, Financial Analyst, Marketing Manager…