Boris Cherny, the creator of Anthropic’s Claude Code, first challenged the old prompt-engineering playbook. He argued that increasingly capable models have reduced the value of lengthy, rigid instructions: “Talk to Claude the way you would a coworker.”¹

His formula is deliberately spare — tell the model what to do, how much effort to spend and how to check its work. Rather than dictating every aesthetic decision in one example prompt, Cherny said he trusted Claude Opus 5.5 to make judgments on its own: “I trust it to cook.”Addy Osmani (addyosmani)² The stance is a marked shift from the era of stock add-ons and prompt rituals: the human supplies the destination; the model is expected to navigate.

Anthropic’s product announcement gives that philosophy a practical, cost-conscious companion. The company says Claude Haiku 5.5 is its fastest model to date outside Opus models running in Fast Mode, and costs “around 75% less” on average than Haiku 4.5.³ Available across all platforms, it is also the first Haiku-class model with an adjustable effort level — a feature that maps neatly onto Cherny’s advice to specify how hard the system should think.

Developers are already framing Haiku as a worker inside a larger AI setup, rather than a stripped-down chatbot. Addy Osmani called it “a cheap, fast and very capable small model,” saying his team uses it as a subagent alongside Opus 5.5.Addy Osmani (addyosmani)⁴ Anthropic’s target is high-volume, price-sensitive work such as summaries, compaction and database queries.

The wider contest remains unresolved. Elon Musk, promoting Grok’s account, urged users to “Try using @Bot for your hardest tasks!”Elon Musk (elonmusk)⁵ Anthropic, by contrast, is making its case from both ends: cheaper automation for routine volume, and less ceremonial prompting for complex work.