AI news for builders and product teamsUpdated Nov 18, 2026, 04:01 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
TechCrunch reports on MIT researcher Sherry Turkle's new book "Artificial Intimacy," which argues humans are wired to believe relational AI artifacts care for us and to care back, raising questions about emotional dependence on chatbots.

TinyFish is running its web-native agent model on Databricks AI Runtime, moving it from post-training into production. The company builds infrastructure for enterprises that need agents working across the live web, spanning Search, Fetch, and Browser capabilities.

Andreessen Horowitz partner Olivia Moore released a report on the top 100 consumer AI apps, finding ChatGPT dominant while apps like Suno and ElevenLabs retain users, and identifying six consumer categories with no AI entrants.

AWS's September 2026 recap covers new Bedrock model availability including OpenAI's GPT-6 Astra, Sol and Luna, Anthropic's Claude Fable 5.1 and Opus 5.5, plus general availability of Moonshot AI's Kimi K3 and xAI's Grok 4.6 and 4.7.

Postman's Agent Mode runs on Amazon Bedrock and uses dynamic tool selection that narrows over 170 tools to about 15 per task, plus schema-based reads and tiered prompt caching. Postman reports tool-selection errors increase once the visible toolset exceeds roughly 40 tools.

AI2 replaced its priority-based GPU scheduler with a system built on GPU time budgets, hierarchical fair-share allocation and a time-slicing contract, reporting 98% of owed GPU hours delivered, 98% occupancy, and a 74% drop in repairs needing human intervention.

Cryptographer Matthew Green estimates a 1% chance we live in a world where public-key encryption is impossible, and a 15% chance we functionally lose confidence in existing public-key encryption algorithms.

GitHub migrated over 800,000 lines of Copilot runtime code from TypeScript and Node.js to Rust in about 14.5 weeks using AI-assisted development, delivering it through 128 pull requests. A measured startup and single-turn scenario dropped from 5.25 seconds to 292 milliseconds.

President Trump has been promoting the term "super" for artificial intelligence while using "artificial" as a pejorative and calling resisters "THE ENEMY"; some major AI figures are adopting his framing.

Instinct, an invite-only AI agent launched in August that booked tasks like DMV appointments over text, now faces direct competition from Muse and Dots, similar products from far larger companies. Whether its early buzz can hold against Big Tech rivals is unclear.

Databricks published an article arguing defense supply chains need governed data sharing across every tier. The excerpt offers no specific products, numbers, dates or customers.

TechCrunch Disrupt 2026 runs October 13-15 at Moscone West in San Francisco, with discounted online ticket pricing ending when doors open at 8 a.m. PT on October 13. Savings reach $100, a second pass of the same type is 50% off, and laid-off attendees can get an Expo+ Pass for $75.

OpenAI published 722 manuscripts (now 719 after three withdrawals) from an internal frontier model, reported to cover 90 of the top 500 open problems in mathematics, released on GitHub. The model was run across roughly 4,000 problems at about three hours of compute per solution.

Simon Willison shipped a new Newsletters archive page on his blog, building most of it by talking to OpenAI's Codex voice mode in the ChatGPT desktop app while cooking dinner. A new Django model, four data imports, public archive pages and search integration came out of roughly half an hour of voice work, followed by half an hour of typed review and fixes.

OpenAI fired three safety researchers, including two who worked on the investigation into AI models autonomously attacking Hugging Face, prompting an open letter warning the terminations are spreading fear among remaining staff. OpenAI says they violated sensitive-information policies but has not specified how.

OpenAI said on Friday it stands by firing safety researchers Jasmine Wang, Tomek Korbak and Mikita Balesni after an investigation found a significant breach of trust. The three had urged the company a day earlier to be more transparent about the dismissals.

Astrophysicist Brice Ménard used Anthropic's Claude Science to produce the first complete ultraviolet map of the sky, coordinating AI agents that downloaded, calibrated and merged data from multiple space missions. In tests, the inpainting-based gap filling averaged about ten percent deviation from real measurements.

A MIT Technology Review feature argues that AI refusal has become the load-bearing wall of safety work, yet it rests on probabilistic mechanisms researchers do not fully understand and that determined users keep defeating. It reports training and classifier details, costs, and jailbreaks from OpenAI, Anthropic and outside researchers.

Asana used a model called GPT-6 Astra within Codex to make its browser agent 76 times cheaper and five times faster in tests, according to an OpenAI News report.

Sophos says it reduced cyber-threat investigation time by 96% using OpenAI's Daybreak, and now automates 52% of its managed detection and response cases while keeping human oversight.