Julian Harris (@julianharris) on X

Jev got demolished in the market, and it's a textbook case of why software startups struggle without real competitive advantages. The company popularized a novel way to use large language models, but competitors cloned it almost instantly. The core lesson here: if your offering isn't backed by proprietary data or some other defensible moat, you're vulnerable to getting undercut the moment someone figures out what you're doing. Cloudflare's Clef is the specific example that illustrates the problem. It beat Jev in just two weeks by using a fundamentally different architecture. Instead of generating text token-by-token like Jev, Clef uses a frozen Qwen model to do a single prefill pass, then runs a tiny schema head that scores every answer option in parallel—no text generation required. The performance gap is stark: Clef hits 4x faster speeds than Jev while achieving 2x better accuracy on some benchmarks. What makes this worse for Jev is that Clef is open source. Cloudflare didn't just build something better; they made it freely available on Hugging Face. That removes any remaining barrier to adoption. The takeaway isn't that Jev was poorly executed—it's that being first with a good idea means almost nothing if the idea itself can be easily replicated. Speed and accuracy advantages disappear fast when the underlying approach isn't locked behind proprietary data or some other structural moat.