Opinion · SURFACE editorial · Intellectual journey series · Public narrative · Receipt-backed where cited
Honesty stamp: GREEN
So what is still open?
Today the hard question is different. It is not whether a model can learn from data, or whether a graph can survive a node failure.
It is whether agents — software that acts on your behalf — can learn and recover while doing real work, over long horizons, with consequences, and leave behind something a stranger can trust.
Can an agent finish a twenty-step job without drifting? Remember what mattered last week without drowning in noise? When it claims "done," can anyone else verify that — without taking anyone's word for it?
That is the frontier. And it is mostly still open.
What would you need to trust an agent's done?
William broadcasts this beat sparingly and replies heavily — inviting debate, not delivering sermons.
Navigate: ← Part 1 · Part 3 → · Series hub