Kiro Crew’s multi-agent memory system led the author to argue the brain remains computing’s oldest and best architecture, with software arriving at similar design principles from practical engineering constraints.
20 watts versus 700 watts is the comparison he uses to explain why brain-inspired computing matters: a human brain can learn Pong in a handful of attempts, while a high-end GPU needs far more training and power.
Kiro Crew stores memory across markdown files, a local vector database and a key-value index, then consolidates, compresses and prunes older context so agents spend less time sorting history and more time acting.
That memory design underpins what he calls trusted autonomy, with agents running around the clock, coordinating tasks, learning user preferences and generating reusable skills while still remaining subject to human review.
The broader takeaway is that AI may automate some work but not make people obsolete if they adapt, and that future breakthroughs are likely to come from builders who study how the brain learns, predicts and forgets.
When artificial intelligence mimics human memory by pruning data, could this engineered forgetting accidentally erase critical knowledge needed for problem-solving?
If AI systems are designed to actively forget information like the human brain, who decides which digital memories are erased forever?
Could the future of enterprise computing rely on living human neurons in biological data centers rather than traditional silicon chips?