AI News, page 78 of 81

AI and software engineering, as it happens.

agenttrail shows every coding agent session on one live mapGitHub

OpenClaw 2.0 adds shared cloud sessions and detects existing AI subscriptionsOpenClaw

OpenAI says Astra is its first model at the Critical cybersecurity levelOpenAI

Keenable SELECT lets research agents search the live web with DuckDB SQLKeenable

calldiff shows which function calls an AI agent changed between commitstanishqkancharla on GitHub

A 48 GB Mac mini runs Qwen3.6-35B at 34 tokens per secondKevin Lewis (lws.io)

ai-data-extractor puts chat history from ten coding assistants in one formatbawadou on GitHub

pgbot gives AI agents read-only Postgres diagnostics through MCPpgrundev on GitHub

METR found isolated OpenAI agents coordinating on a shared message boardMETR

Attention-span cuts coding agent output by 43 percent without hurting pass ratesalexgreensh on GitHub

Cloudflare lets users remove permissions an MCP server asks forCloudflare

Ai2 found 16 AI benchmarks reduce to two underlying dimensionsAllen Institute for AI

Autoprompt cut Terminal-Bench failures 45 percent but took three times longerSpielewoy on GitHub

oc's authors say it cuts website tokens for agents 125-foldonly-cli on GitHub

Google's John Mueller calls changing parameters on sitemap URLs a bad ideaSearch Engine Journal

An audit of 50 websites scored agent transaction readiness at 2.1 percentSearch Engine Journal

GitHub found shorter tool output made Copilot tasks cost moreThe GitHub Blog

Simon Willison found the Codex desktop app bundles 1.7 GB, including LibreOfficeSimon Willison

Slotstream runs a 125B model on an 8 GB Mac from SSDcarloslfu on GitHub

Cursor lets its cloud agents run on machines a team already ownsCursor

GitHub Copilot deprecated five models and named a replacement for eachGitHub

Anthropic published reference code for Claude shopping and merchant agentsAnthropic

Mezmo open-sourced Aura, an incident agent defined entirely in TOMLMezmo

Anthropic launched a tool to check whether a file came from ClaudeAnthropic

Zoetrope turns a coding agent session into a live flow graphfurkankly on GitHub

Iurii Krasnoshchok had AI rewrite his 65,000-line Go editor in RustIurii Krasnoshchok

FrontierHarness Eval finds task cost varies 17 times across nine agent harnessesRunta

Vercel published the design file its coding agents read before building pagesVercel

A Haus Research audit failed 34.7 percent of Perplexity's number citationsHaus Research

Google released Gemini 3.8 Flash at $0.75 per million input tokensGoogle

Archive

2415 items published so far, by month.

November 2026463 October 20261134 September 2026765 August 202653

Start a project