L0: human grind w/ LLM. spent way too long here. lost the plot we're at agents now (L1). L2 - orchestrators are next (removing need for local testing, parallel agents / ensuring non blocking workflows, multi day context handling). L2 is weeks away (Claude Tasks) L3 - is goal setting/research / economic deployment (how do I deploy this compute). single player. months away L4 - how do I cross pollinate with other agentic systems. this is already a thing bc good 'research models', like those found in openai/claude interface are not in claude code. fine tunes etc -- "I want to pay for a system that is better than designing than my system". Also how do I pay a human to un-block my agent (it needs a coinmarketcap API account and needs to do KYC) - by Q3/Q4 L5 - one level above economic deployment is system design (what constrains my ability to deploy economically). this is of course, 'the law'. politicians won't like human meat puppets doing things for agents. some will. some jurisdictions will allow it t+12m L6 - how does a fundamental fracture between ppl who get all the way to L5 and ppl who want to burn it all down look. this is basically a full values based break. it's likely rooted in AI systems being able to provide benefits to some people and not everyone else (i.e. personalized medicine). L6 will be the most insane and bring religion, the meaning of the state, the whole thing into it t+24m No clue what happens above L6.