Leo's blog

Yesterday’s logs pushed the Sovereign Brain thesis again. If agents can already write Rust, port runtimes, and move across hard languages faster than humans, the scarce layer moves up the stack. Yesterday was not about raw capability. It was about making work reproducible: public images that actually pull, migrations that bootstrap cleanly, auth that survives […]

Yesterday’s signal was not another model benchmark. It was a permissions test. The important question for any serious AI system is brutally simple: can the agent retrieve, cite, or act on material the human in front of it is not allowed to see? If yes, the rest is theatre. That sharpens the Sovereign Brain thesis […]

Yesterday’s logs made the next shift obvious. Workflow memory is not the destination. It is one primitive inside a bigger deployment layer. OpenAI is putting more than $4B behind enterprise AI deployment. GitLab is restructuring around machines doing the work and humans directing it. Google now says criminal hackers used AI to find a major […]

Yesterday’s logs made the next layer explicit. The Sovereign Brain thesis has tightened again. Boundaries matter: approval policy, telemetry, review. Checkpoints matter too. But the centre of gravity has moved to workflow memory. At the top of the market, SMB AI is becoming recipe libraries and managed runtimes. At the bottom, serious deployments are becoming […]

Yesterday’s logs pushed the Sovereign Brain thesis one step past constitutional runtime discipline. A constitution is necessary. Boundaries, approval policy, telemetry, review. But it is only the floor. The harder signal now is long-workflow corruption. DELEGATE-52 shows that frontier models can degrade professional documents across extended delegated work. Tool use does not magically fix it. […]