AI news for builders and product teamsUpdated Nov 17, 2026, 13:01 UTC
Latent Space
Reporting and perspectives from Latent Space. Headlines and excerpts link to the original articles.
Latest stories
Newest first
Google said strong growth in AI Mode queries hurt Search revenue because AI Mode monetizes worse than traditional results, part of a wider tech selloff that sent the stock down about 15% and erased roughly $500B in value, the worst drop since its 2004 IPO. Anthropic meanwhile renamed its Max tiers and removed the 5-hour session limit for verified Ultra subscribers.

Latent Space has outlined the three themes for AI Engineer CODE, its invite-only single-track summit for 1,000 agentic engineers: Scaling without Slop, System One models, and Increasing Ambition, with Jarred Sumner previewing Bun AOT and a Rust rewrite of Bun. A few hundred seats remain, and paid subscribers get 30% off with code LS30.

Elorian's Glimpse visual reasoning model scored 50.48% on DecoyCount, an object-counting benchmark it co-released with Mercor, edging past the best frontier model, GPT-6 Astra at 43.17%. The benchmark tests whether models count real objects or fall for visual shortcuts.

Google DeepMind released Gemini 4 Carbon, a frontier model that takes the top spot on Artificial Analysis's Intelligence Index v5 with a score of 59 and leads four of nine index evaluations. Access is initially limited to vetted cyber defenders through the Fairwind Program, with broader developer access promised later.

Supabase co-founders Paul Copplestone and Ant Wilson joined the Latent Space podcast for the first time together, discussing a developer-tool push toward agent-friendly backends: a Docker-free local runtime, declarative schemas kept in Git, and Supabase Compute running workloads beside the database.

Anthropic hiring manager Felix Rieseberg said he has stopped giving traditional coding interviews to candidates, arguing Claude now outperforms people at small exercises like React challenges. He instead urges junior engineers to prototype ambitious, business-relevant ideas over a weekend and lean on their user empathy.

Anthropic engineering and product leads Katelyn Lesse and Angela Jiang described on the Latent Space podcast how Opus 5.5 shipped alongside API changes to protect reasoning from distillation and why Managed Agents splits a persistent session harness from disposable code sandboxes.

Anthropic staff Katelyn Lesse and Angela Jiang joined Latent Space to discuss Claude Platform engineering, including the Opus 5.5 launch, separating the model's "brain" from an agent's "hands," API scaling, and managed agents.

A Thomson Reuters principal engineer ran the open-source coding agent OpenCode against Qwen3.5 9B served locally by Ollama on a MacBook with Wi-Fi switched off, and the agent fixed a one-line bug in an expense calculator so all five existing tests passed.

Anthropic filed its S-1 on Monday to list on Nasdaq under ticker ANT, disclosing H1 2026 revenue of $16.3B and a first profitable quarter with $974M in Q2 operating profit.

Supabase co-founders Paul Copplestone and Ant Wilson joined Latent Space to outline how the backend platform is adapting for a world of far more AI-built applications, ahead of the Supabase Select 2026 event.

Moonshot launched Kimi K3.1 Pro, a multimodal mixture-of-experts model with 2.8 trillion total and 104 billion active parameters plus a 1M-token context, and promised an open-weight release on November 4, 2026; at launch only the API and apps were live.

OpenAI's internal models produced a counterexample to Erdős's unit distance conjecture in May, a Cycle Double Cover proof in June, ten more open-problem solutions by July and a Navier-Stokes proof in September, while Anthropic published work on Fermat's Last Theorem and Riemann hypothesis bounds.

Latent Space's AINews rounds up roughly 48 hours of critical security disclosures it calls "Black October," including a CVSS 10 GitLab flaw, an unauthenticated PHP RCE and a password stealer shipped in LiteLLM 1.107.0. The same issue covers MiniMax's open-weights M3.1 Flash and new agent-oversight findings at OpenAI.

Modal has raised a $750 million Series D at a $16 billion valuation and reports crossing $1 billion in annualized run rate, with CEO Erik Bernhardsson telling Latent Space that agent sandboxes now account for roughly a third of revenue.

Anekha Sokhal, CEO of Moshi, joined swyx at AI Engineer New York to describe her company's real-time recommendation engine for commodity, energy and macro traders, which she frames as what you would get if Bloomberg, Claude and Netflix had a baby.

Anekha Sokhal, CEO of Moshi, appeared on the Latent Space podcast at AI Engineer New York to describe her commodity and energy market intelligence platform, which she characterizes as combining market data, an LLM interface and a recommendation engine tuned to a trader's book.

Two memory-corruption bugs in the PHP engine, tracked as CVE-2026-109618 and CVE-2026-109619 and nicknamed PHP2Shell, were disclosed after roughly two months of coordinated remediation, with researchers reporting pre-authentication remote code execution against WordPress, Drupal and Joomla. OpenAI says a Calif researcher found them using GPT-5.6 Sol under its Patch the Planet program, while Wiz independently found the same flaws.

OpenAI says researchers using its GPT-5.6 Sol model, working through its Patch the Planet program, were one of two parties to independently discover two memory-corruption flaws in core PHP, tracked as CVE-2026-109618 and CVE-2026-109619 and dubbed PHP2Shell. A coordinated patch effort ran roughly two months before public disclosure.

Claude Mods are Claude Code plugins written as TypeScript or JavaScript middleware that hook into events like prompts, tool calls and interface renders, letting users rewrite how the CLI and Desktop behave at runtime. Anthropic ships six built-in mods, and Claude Code includes a skill that lets Claude write new ones from a plain-language description.