AI NEWS · UPDATED DAILY
What changed in AI this week.
New releases from the companies themselves, what builders are discussing on Hacker News, and repos and skills worth a look. Picked each morning by a scoring script, at most five stories a day, and every link goes to the original source.
2026-09-24
@ClaudeDevs: Try these in your first Opus 5.5 session: → Hand over a whole task. Define "done" and when to check in. →…
Try these in your first Opus 5.5 session: → Hand over a whole task. Define "done" and when to check in. → Drop "think carefully". It always thinks first. → After a long run, check what it needs to go further. Our playbook:
Ringg’s AI agents resolve up to 65% of customer calls with OpenAI
Using GPT-5.6, Ringg powers multilingual agents across voice, chat, WhatsApp, and web for 90% less cost vs. GPT-4.1.
2026-09-23
Claude Code templates for agents, skills and MCPs
A CLI catalog for Claude Code agents, commands, hooks, settings, skills and MCP integrations, with installable templates and monitoring tools. Useful for teams standardizing agent setups.
Univer office SDK for AI-assisted products
Univer is an open-source SDK for embedding spreadsheets, documents and presentations in products, with browser and Node.js support. It fits teams building AI-assisted office or data workflows.
Google AX agent orchestration runtime
Google’s AX is a Kubernetes-shaped runtime for sandboxed agent tasks, workspaces, MCP servers, model settings and network gateways. Useful for teams that need repeatable, isolated agent workloads.
Claude for Financial Services
Anthropic’s reference repo packages finance agents, skills, commands and MCP connectors for research, modeling, deal work and operations. Useful as a starting point for governed internal workflows with human review.
Claude Code changelog adds model, MCP and agent fixes
Claude Code’s latest changelog lists Opus 5.5 support, a larger MCP description limit and fixes across hooks, symlink writes, subagents and long-running sessions. Useful for teams updating agent setups.
Claude Opus 5.5
Anthropic’s new Opus model targets long-running coding, computer-use and knowledge-work tasks, with lower token pricing than Opus 5. It is useful for builders comparing agent costs and capability.
AI coding has made CI a bottleneck, so we reworked ours to keep up
Linear explains how faster AI-generated changes made CI the bottleneck, then details runner, checkout, setup, sharding and test-execution changes. Useful for teams scaling agent-written code.
Orchestrating Claude Code agents with a chief of staff pattern
Asyncdot describes a coordinator-worker pattern for long coding jobs: durable task state, separate execution sessions and re-run verification. Useful for builders managing multiple agents without trusting self-reports.
Life Sciences Verification Program
Anthropic’s beta program gives vetted life-science teams access to Claude models with biology-specific safeguards and review. It helps research, biotech and pharma teams test broader workflows responsibly.
Grok Bot handles customer support at SpaceXAI
SpaceXAI says it rebuilt support around Grok Bot, connecting ticketing and issue tracking to investigate tickets, suggest replies, flag incidents and prepare reports. Human approval was used for writes during rollout.
xAI releases Grok 4.7 for coding and knowledge work
Grok 4.7 targets longer coding and office tasks, with better self-checking and context handling. It is available in Cursor, Grok Build, the Grok API and other model platforms.
Jev becomes the fastest-adopted model on Vercel AI Gateway
Jev is a probabilistic decision model for software workflows. Developers can use its typed choices and scores to route agents, retry or stop jobs, score risk, and send uncertain cases to human review.
xAI releases Grok Voice Transcribe 2.0
The new speech-to-text model improves noisy, multilingual and multi-speaker transcription, with streaming, timestamps, diarization and key-term controls. It keeps the previous API price and will replace version 1.0 as the default.
Vercel reports open-weight models reached 56% of AI Gateway tokens
Vercel’s September AI Gateway index says open-weight models handled 56% of August token volume, while average token prices fell 23.2%. The report helps builders compare model adoption and production costs.
Claude Projects now coordinate multiple Claude Code threads
Claude Projects can set a goal, delegate work across parallel Claude Code sessions, share project memory and files, run tests, and open pull requests. The beta is rolling out first to select Pro and Max users using cloud sessions.
LETTERS FROM MY DESK