Cognition, a six-month-old startup in the AI coding space just raised $175m at a $2b valuation:

Despite the skepticism surrounding Devin’s launch, the AI coding assistant has shown promising results. According to the SWE-Bench benchmark, which evaluates AI models on software engineering tasks, Devin achieved a 13.86% accuracy in resolving issues unassisted, surpassing the previous best model’s 1.96% unassisted accuracy.

I am (to put it mildly) very skeptical of their ability to defend that valuation, but I wanted to zoom out a bit and offer a small bit of advice when it comes to understanding the potential and value of tools like these:

You need to be able to taste the kool-aid they're selling you.

I feel like I am in a very weird uncanny valley of bearish on the current crop of AI-enablement tools and yet very very bullish both on the ones that I use (which is to say, Copilot) and the overall potential for the industry going foward. Why? Because every single tool I use tends to be bad, and bad in obvious ways that make me suspicious of demos.

I am quite excited for Cognition to open up their beta so I can try it out — I might eat my words, and certainly if any company can, say, double the efficiency of a median developer than they've earned a multi-billion dollar valuation. But until then, I assume that it is vaporware whose differentiation will be gobbled up by players (Microsoft, OpenAI, Sourcegraph) more well-positioned to execute at scale.