I spent part of May testing OpenClaw workflows, and the thing that stayed with me wasn't speed. It was continuity.
At first, I assumed better automation was mostly about better models. Smarter outputs. Higher benchmarks. Cleaner responses.
Then I watched a workflow run for nearly 19 hours.
What stood out wasn't what it completed. It was what it remembered.
A lot of AI systems still feel surprisingly fragile once time enters the equation. They handle Task A well, but by Task B or C, context starts leaking. The user becomes the memory layer.
OpenClaw didn't eliminate that problem. I still saw edge cases and drift. But the failures felt more environmental than cognitive.
That made me wonder if the bottleneck is shifting.
Maybe reliable automation isn't primarily an intelligence problem anymore. Maybe it's a continuity problem.
Still early, obviously.
But if AI can act for days while preserving context, the real question becomes who owns the systems that remember, and how value flows through them when they do.