Every part of an agent harness is a bet the model can’t do something yet. Anthropic keeps finding the bets expire on upgrade, and the skill is knowing which just did.
Partly, though I'd put the moat one layer down. Context engineering that compensates for what the model can't do yet is a harness component, and those expire on upgrade: this year's clever retrieval is next year's dead weight. What doesn't expire is having data nobody else has, and a loop that keeps generating more of it. The plumbing gets commoditized every generation. The well doesn't.
Awesome, I completely agree with you. Accumulating unique assets early on is what creates a true advantage. I’m honestly getting tired of seeing endless daily posts about new frameworks and harness techniques, once the core intelligence becomes strong enough, all that engineering will be abstracted into a black box, while whatever is tied directly to the user remains genuinely unique
Do you think context engineers are what really determine a competitive moat because they deal with a company's proprietary data and unique knowledge?
Partly, though I'd put the moat one layer down. Context engineering that compensates for what the model can't do yet is a harness component, and those expire on upgrade: this year's clever retrieval is next year's dead weight. What doesn't expire is having data nobody else has, and a loop that keeps generating more of it. The plumbing gets commoditized every generation. The well doesn't.
Awesome, I completely agree with you. Accumulating unique assets early on is what creates a true advantage. I’m honestly getting tired of seeing endless daily posts about new frameworks and harness techniques, once the core intelligence becomes strong enough, all that engineering will be abstracted into a black box, while whatever is tied directly to the user remains genuinely unique