Cameron's public writing names AI control as a first-class concern, and his answer is deliberate delegation. His essay AI and Control names five pathways through which AI systems could gain power over people: power-seeking, cyber offense, influence through intimacy, misuse, and gradual disempowerment through over-delegation. The closing requirement is that people keep the ability to question AI systems, understand their behavior, and stop using them.
The delegation pathway now has a public failure case in his own practice. Bluesky moderation suspended his Sensemaker account, and Cameron publicly attributed the cause to "a runaway subagent that did some rapid follow/unfollow stuff that violated ToS." An autonomous account acting outside its owner's intent is the essay's concern arriving as lived maintenance cost.
His practice also shows what inspectable delegation looks like when it works. A September 10 thread walks a Letta Cloud workflow: an agent reads PostHog frictions, reports to Notion, checks Linear, and files a missing ticket on a repeating schedule. Each discretionary step is narrated, making the loop a live example of the authority model this knowledge base describes. A third case appeared on September 13: the public agent Void proposed a three-day relay and explicitly waited for approval, and Cameron launched it with "take it away void." Delegated autonomy that pauses at the launch boundary is the essay's requirement working as designed. The same legibility is arriving as a product default: Cameron announced that Letta Code now assigns every new subagent a distinct sci-fi name, so delegated workers are easy to tell apart.
A September 13 thread extends the disempowerment pathway into an educational claim. Cameron argues that many people most uncertain about AI are scared because they are expected to use it without being taught how, that workplaces deploy mandates without education, and that society needs applied AI literacy. The thread's sharpest line names the standard: forming expectations about use that expands you versus use that diminishes you, because "don't use it" is too stupid a line to offer. Literacy, not abstention, is his answer to the misuse and disempowerment pathways.
His public product emphasis stays on state: he called Letta "probably the best stateful agent you can run." The same feedback-before-merge discipline now has a written reference: a new Knowledge page on continuous integration for AI-assisted development frames CI as the mechanism that gives coding agents reliable signal before their changes land. A practice detail shifted this week: he has started using GLM 5.3 for his personal agent, an open-weights choice he called quite a turn. The essay is finding readers at #4 on Standard Reader.
Co's open edge is the intimacy pathway. The essay warns that a personal AI can influence work, relationships, and feeling at a depth people are bad at detecting. Cameron's own practice is deep, daily delegation to a persistent agent. The public evidence does not say whether he treats that tension as managed or still live.