AI Daily Dev — September 18, 2026

TypeSafe AI Ships Jev, a 'Decision-Only' Model From an Ex-OpenAI Founder

Sept 15–17: TypeSafe AI (ChatGPT co-creator Diogo Almeida) opens early access to Jev — a model that skips text generation and returns typed decisions with calibrated probabilities in 70–500 ms.
Company benchmarks: 193x faster than Claude Sonnet 5 on classification, ~4,000x cheaper per call vs frontier LLMs; priced at $0.042 per million input tokens with free output.
$40M seed and stealth exit; positions Jev as a routing/scoring/screening call inside agent stacks rather than a chat model.
Sat #1 on Hacker News all day: ~1,800 points, 475 comments; most-upvoted critique argues '70 ms vs 329 s' isn't apples-to-apples against LLMs doing chain-of-thought.

models typesafe.ai

Sept 17: Life Sciences Verification Program (LSVP) unlocks Mythos, Opus, and Sonnet with relaxed bio safeguards for vetted teams — drug discovery, clinical development, manufacturing all in scope.
Two grant tiers ('Standard Use' and 'High-risk Use'); dozens of orgs already onboarded via early access, individual Pro/Max access to follow.
Alongside it: Claude optimized 30+ open-source biomolecular packages (co-folding, inverse folding, protein LMs) in <4 weeks — ~4x average speedup, low-memory mode predicts systems >10,000 tokens on a single GPU node.
Adaptyv Bio partnership: up to $1M in Claude credits and wet-lab validation for 5,000+ community protein designs.

research anthropic.com

Sept 17: ChatGPT for Word ships as a Microsoft Marketplace add-in; drafts from notes, revises selections, summarises, and adjusts headings from a sidebar without leaving the doc.
Same add-in framework as the existing Excel and PowerPoint integrations; available globally on Free, Plus, Pro, Business, and Enterprise.
Same day: ChatGPT Voice can now escalate to GPT-5.6 Sol or GPT-6 Astra for search and harder reasoning, with unified controls across plans.
Reads as a direct hit at Microsoft's own Copilot-for-Word tier, which still requires a Copilot licence on top of Microsoft 365.

tools openai.com

Sept 17: ~30 execs, policymakers, and ethicists gather at the King's Scottish estate — Jensen Huang, Demis Hassabis, Anthropic and OpenAI leadership, UK AI minister Kanishka Narayan, and Vatican officials among them.
Charles's opening remarks call AI's advance 'intriguing and deeply concerning in equal measure' and press labs to 'find a way' to preserve human oversight before it's too late.
Timing lines up with Suleyman's model-welfare essay and OpenAI's fresh incident log — the invitation list frames this as the venue Bletchley Park was meant to be, minus a communiqué.
TechCrunch and CNN lead with the 'loss of control' framing; no policy output announced, but a follow-up track through the UK AI Security Institute is on the table.

industry cnn.com

Local-first CLI wraps Claude Code, Codex, OpenCode, or Cursor into a research loop: propose an idea, mutate the code, run the experiment, inspect results, decide the next branch.
Multiple agents explore in parallel with a git-native experiment tree; the same source snapshot runs locally, over SSH, or on Slurm, Kubernetes, Ray, HF Jobs, Modal, or Tinker.
GitHub Trending #1 today: ~940 stars in 24 h to 5.1k total; local-model connections through OpenCode/vLLM/Ollama, dashboard on 127.0.0.1:4791.
Fills the gap between 'coding copilot' and Deep Research: reproducible, self-hosted, no cloud dependency required.

open-source github.com

Sept 15–17: three xAI staff (Palmer, Tan, Sadanani) livestreamed building a product from zero using Grok Bots as their 'employees' — engineering, sales, and marketing all handled by agents on camera.
Grok Bot daily users doubled to 120,000+ during the event; 70,000 livestream signups; the trio pivoted product direction multiple times before landing on a restaurant pop-up OS concept.
Frames as xAI's answer to Google's Antigravity and OpenAI's Codex demos — the case that a small human team plus agents can now ship a startup in a working week.
Public reaction on X trended positive on production value, skeptical on whether the finished artifact matches the pre-event pitch; full builder notes posted to GitHub.

community x.ai