Refreshing counterweight to the "18-month window" panic.<p>The "constancy anchor" framing resonates. I've been building heavily with Claude Code lately and the bottleneck really has shifted from implementation to intention and judgment. The code writes itself; knowing what to build and why is the hard part now.<p>That said, I think the piece undersells how quickly the goalposts can move on what counts as "intention" and "taste." A year ago people said LLMs couldn't do system design. Now they meaningfully can, with the right prompting. I wouldn't bet the farm on any specific human capability remaining exclusively human on a long enough timeline.<p>But the core point (stop panicking, go take a walk, figure out what you actually care about) is probably the most productive advice anyone in tech can hear right now. The people I see thriving with these tools aren't the ones shipping the fastest; they're the ones who had a clear point of view before the tools showed up.<p>The "embodied intention" line is quietly the strongest point in the whole piece. For the constancy anchor to expire, an agent would need something we can't even define in ourselves yet. That's not a gap that closes in 18 months. It might not close at all.<p>The dig at the artifact-bragging-to-impressed-bosses pipeline is chef's kiss accurate.