As AI generates code faster than humans can write it, good taste in programming moves upstream—from judging individual lines to asking the right questions about what to build. The programmers who'll get most out of AI tools are those who already know what they want.

• Human taste during coding happens at typing speed, which constrains damage from bad decisions. AI removes that slowness, so taste becomes a reviewing filter—but reviewing is harder than writing and happens too late to catch fundamental design mistakes.
• The most important choices (what to build, how to split it, what to omit) happen before the model writes anything. A well-executed answer to the wrong question is still wrong.
• Models will get good at local code judgment (idioms, naming, structure) but can't learn domain-specific taste—knowing which complexity matters, how users will behave, where the system will change. That knowledge lives with the people who own the problem.

Jared Palmer released Kev, a family of four open-source models (0.8B to 27B parameters) designed to answer multiple-choice and yes/no questions by scoring answer options directly rather than generating tokens. The models run on consumer hardware, come with Apache-2.0 weights, and match or exceed the performance of TypeSafe's proprietary Jev API.

• Kev-27B achieves 75.7% accuracy on generalization tasks and 91.8% on policy reasoning, with smaller variants (Kev-4B, Kev-9B) performing within 1-2 percentage points on specific workflows while using far less compute.
• The architecture uses a pointer head to score supplied answers without token generation, and caches document computation across multiple questions—cutting inference time from 9.4 seconds for a new 64k-token document to 0.7 seconds per follow-up question on an H200.
• Kev-4B runs on a 32GB Mac (136ms per question with cached text) and Kev-0.8B on any Apple Silicon, making local deployment and fine-tuning practical; the largest model costs $44.09 per million requests on H200s versus $3.54 for the smallest.