Zig, Zed, Anthropic: when a language creator calls the hype by its name

A scathing post from the creator of Zed calls out Anthropic on its agent promises. The language: "Anthropic blows smoke." The 2026 hype fatigue is hitting founders we didn't expect to see on this battlefield.

In plain terms

A scathing post published on July 13th - titled "Zig Creator Calls Spade a Spade, Anthropic Blows Smoke" - calls out Anthropic on its agent promises. The text, hosted by Ray Myers, directly targets the gap between the marketing demo and the production experience. Whether the post is more specifically about Zig or Zed, the mechanism is the same: an insider from the tooling world breaks the silence.

The context

The 2026 hype fatigue had an identified profile: geohot, Karpathy in some interventions, Lambert on Interconnects. Side voices - independent, opinionated, non-corporate. What this post marks is the arrival of another type of voice: that of actors directly dependent on the AI stack. The creators of code tools - whether Zed the editor or Zig the language - sell a developer experience that relies on LLMs. Publicly stating "blows smoke" is not gratuitous.

What the post says

The title is enough to frame the thesis: Anthropic "blows smoke" about its agent capabilities. The crux of the criticism - not detailed in the summary accessible on the KEEL CRUX side - revolves around the reproducibility of agentic demos outside laboratory conditions. It's the classic complaint of 2026 hype fatigue: the demo works, the production workflow fails.

The analysis

What matters is not the specific technical complaint - everyone has their own. What matters is who is speaking. A founder of AI-first code tooling, whose business relies on LLMs working, publishes "it doesn't work as advertised." The hype-fatigue thread shifts from observer frustration to insider speaking out. It's the end of the free pass on agentic demos.

So what

For Anthropic (and OpenAI): agent capability benchmarks must include publicly reproducible workloads, otherwise credit erodes quickly. For a decision-maker driving agent adoption: ask for the unedited raw video of the workflow the provider promises, not the demo. For an investor: the "demo → prod" layer is the real 2026-2027 gap; those who bridge it will disproportionately win.