Summary
The Cognitive Revolution’s AI:AM weekly highlights show compiles clips from three live sessions covering some of the most consequential AI developments of the week. Co-host Praash Naran reports spending an entire weekend running three to four simultaneous Astra agents on his production codebase, concluding that GPT-6 Astra has crossed a threshold of genuine usefulness—resolving long-standing bugs, performing reliable computer use, and completing tasks like labeling 12,000 basketball-court images that previously required human labor. Naran goes further, describing Astra as AGI in a practical sense: capable of doing things better than most hirable, trainable people.
A key technical discussion covers how Astra manages extended context: rather than compacting million-token sessions into summaries that discard detail, the model maintains a persistent notes file and can search its own session history, effectively multiplying usable context by roughly ten times. The hosts also note that Meter’s famous “hours of work” benchmark chart has become functionally obsolete—model development cycles are now shorter than the time required to design tasks long enough to measure at current capability levels.
The episode also addresses Sam Altman’s statement to OpenAI staff that he is open to pacing frontier model development, contingent on competitor coordination, and features a substantive debate about human agency versus technocapitalism with guests from Mozilla’s Mythos project examining whether voluntary industry agreements or government enforcement is the more realistic path to meaningful AI governance.
📺 Source: Cognitive Revolution “How AI Changes Everything” · Published September 12, 2026
🏷️ Format: Podcast
