Articles.
Field notes from the studio on applied AI, typed model judgments, and the tools we use to ship them into production software.
7 min read
What is Jev? TypeSafe's System One Decision Model Explained
Jev answers typed questions about your data with calibrated probabilities, and never writes a word of prose. What it is, how it differs from an LLM, and where it fits in real software.
Read the article6 min read
Choice, Score and Noul: The Three Question Types
Every question you ask Jev is a Choice, a Score or a Noul. Here is what each one returns, how to pick between them, and how to compose them into decisions your code owns.
Read the article6 min read
Decision Models vs LLMs: When Not to Generate Text
Most AI features need a decision, not a paragraph. How decision models like Jev differ from LLMs, and a practical test for when to stop generating text.
Read the article5 min read
Confidence vs Probability: When to Let Jev Act and When to Escalate
An AI answer is only half a decision. The other half is knowing when not to trust it. How Jev separates probability from confidence, and how we set the thresholds.
Read the article6 min read
Speculative Fan-Out: Ask Every Question in One Call
Ask every question you might need, in one request, and let code ignore the rest. Why that is cheaper and faster than branching, and when a second call is genuinely required.
Read the article6 min read
Jev on OpenRouter: The Decisions API, Pricing and Latency
OpenRouter serves Jev through a separate decisions endpoint, not chat completions. Here is the request shape, what it costs, and how fast it answered for us.
Read the article6 min read
How We Built a Claude Code Model Router with Jev
Not every message to a coding agent needs the biggest model. How we used Jev to size each request and hand small jobs to Haiku and Sonnet, including the limits we hit.
Read the article7 min read
Picking the Right AI Agent Skill with Jev
Agents pick skills from a cramped one-line index and often guess wrong. Here is how we put Jev in front of that choice, and what it costs.
Read the article7 min read
AI Guardrails and Citation Checks with a Decision Model
A guardrail should hand you evidence, not make your policy for you. How we use Jev to screen LLM traffic and catch citations that do not hold up.
Read the article6 min read
Triage at a Fraction of a Cent: Sorting Emails, Leads and Tickets with Jev
Sorting a pile of text is a decision problem, not a writing problem. Our Jev triage setup, what one email cost, and the rules we use when Jev is unsure.
Read the article
Want this thinking on your roadmap?
If an article describes a problem you are facing, send a short brief. We will tell you what we would build, and what we would not.
Start a conversation