This week's briefing (9–15 Nov)Data updated 6:35 IST, 22 NovHow we check claims
Diogo Almeida
Andreessen Horowitz · TypeSafe AI · Builder-founders · Investors. 3 sourced entries, newest first. Curated coverage with primary sources, not a complete feed; a role establishes context, not authority.
TypeSafe AI · 14 October 2026 · Company statement
Jev: For the builders AND dreamers (Jev 1.14.0 release)
Primary company statementModelsStructured outputs
TypeSafe AI announces Jev 1.14.0 plus experimental 'danger zone' features: composable image support (multimodality), a new box type for bounding boxes, and optional longer context, shipped behind separate API flags to keep the production path reliable. The post reports incremental improvements including multilingual support and commits to a deprecation timeline for older versions. No individual author is credited, so it is attributed to the company rather than Almeida personally.
What this could change
Editorial: shows how fast the Jev decision-model category is adding multimodal and longer-context capabilities while fencing them off from production use.
Perspective & interests
TypeSafe's commercial pitch rests on Jev being production-reliable, so this release framing protects its paid API positioning.
Reviewed 2026-11-05. Entry ID typesafe-jev-1-14-release-2026-10 · Permalink
a16z: AI can write code — why isn't software better?
Original publisher metadataAI agentsSoftwareModels
Martin Casado and Ben Horowitz talk with TypeSafe AI founder Diogo Almeida about why software hasn't fundamentally improved despite AI's coding abilities. Per the episode page, coding agents may help write software faster, but the software they produce still largely works the way software always has; TypeSafe's contrasting approach embeds intelligence in the software itself so programs can reason about intent and make probabilistic decisions. The page does not attribute specific individual quotes to Casado.
What this could change
Editorial: the distinction between AI that writes classical software and software that is itself intelligent is a live architectural debate for builders.
Perspective & interests
Casado leads a16z's infrastructure practice and a16z is an investor in TypeSafe AI, the company whose approach the episode showcases.
Reviewed 2026-11-06. Also features Martin Casado, Ben Horowitz. Entry ID martin-casado-a16z-ai-code-software-2026-09 · Permalink
Primary source, read in fullModelsAI agentsStructured outputs
Almeida personally announces Jev, TypeSafe AI's first 'System One Model': an AI that returns type-safe structured decisions with calibrated probabilities instead of generated text, claiming similar intelligence on System One tasks to LLMs while being two orders of magnitude faster (70–500ms vs. seconds), with free output tokens and $0.042 per million input tokens. He describes a new training method, Reinforcement Learning for Calibrated Decisions (RLCD), optimizing for epistemically honest confidence scores, and recounts his prior OpenAI work on instruction-following that fed into ChatGPT.
What this could change
Editorial: a founder's own spec for a non-text, decision-only model class — relevant to anyone wiring AI judgments directly into software logic.
Perspective & interests
Almeida sells Jev API access, so establishing 'System One models' as a category directly drives TypeSafe's revenue.
Reviewed 2026-11-05. Entry ID diogo-almeida-jev-launch-2026-09 · Permalink