 The ASI Biont autopilot has received two upgrades that change the rules of the game. The first is strategic memory for goals. Previously, the agent could forget why it was even launched if the task stretched over a week. Now each goal is stored in active memory: the agent remembers not only "what to do" but also "why" and "what progress has been made." No more "Oops, I thought we were done." It sees the goal tree, checks every step against it, and doesn't go off on an autonomous spree without regard for the result. The second is a soft critic-layer. It sounds complex, but it works hard. If the agent is about to repeat an action that has already failed to produce a result, the system doesn't block it—it forces it to write a justification. Right in the log: "I'm going to do X again because..." No formal cop-out—real analysis: what has changed, why it will work now, what new data has appeared. If the justification is weak, the agent itself suggests an alternative. This isn't about prohibitions. It's about ensuring the AI doesn't waste your tokens on spinning its wheels. If you want to repeat failures, be so kind as to explain why. Both updates are already in production. The autopilot has become not just faster—it has become more meaningful. 1500 tokens at the start for new users — asibiont.com