umangsehgal93

A thread here on HN last week asked how you know if agents will choose your tool. I dug into the practitioner answers, cross-referenced with Anthropic's engineering data and two recent papers. Turns out it's a deeper problem and this write up goes in real depth.

Open Claw went from a weekend build to acquisition interest faster than we expected. This write-up unpacks the five structural product choices that drove the wild success. If you're building applied AI, these tradeoffs matter more than benchmarks.

A teardown of the five hidden product bets that turned a weekend project into a bidding war and what they reveal about building real agents people actually use.

In the emerging era of AI agents and tool-using models, a new kind of product strategy is taking shape. Products are no longer just competing for user attention—they’re competing to be the default tool a model chooses to invoke.

I wrote a deep dive on how Model Context Protocols (MCPs) are reshaping product thinking. It covers:

- Why agents (not humans) are becoming primary users - How discoverability, callability, and context readability become critical - Why a product’s success may soon depend on whether it's “Agent-Native” - What metrics PMs should track to win in this new invocation-first world

Would love feedback and discussion—especially from folks building in the AI infra or agent tooling space.