Welcome back. AI assistants are leaving the chatbox behind, and you get the keys. OpenAI just shipped ChatGPT's voice to its API at a surprisingly low cost. Meanwhile, Google now lets Gemini control your Mac's apps and files.
Also: What connects the Microsoft, airline, and bank outages, Anthropic’s FDE guide, and a programming history video devs can’t get enough of.
Today’s Brief
Lessons from a dev who tried to replace Claude Code
A SpaceXAI engineer’s guide to automating your life
Ship a 21,000-line feature with coding agents (tutorial)
How to orchestrate coding agents (cookbook)

TODAY IN PROGRAMMING
OpenAI hands devs the voice behind ChatGPT: The AI lab just released GPT-Live-1.1, a full-duplex voice model for multilingual conversations through its API. It is the same model Plus and Pro users have used since October 15. It can listen while speaking and is better at picking up tricky details like letters and numbers. A new standby mode kicks in after 5-60 seconds of silence, so you’re not paying full price when nobody’s talking. Active sessions cost $0.05 per minute. Try it here.
Google gives Gemini control of your Mac and tightens Android installs: The search giant’s AI assistant can now handle multistep tasks across your Mac’s apps and files while you get on with other work. It’ll even wait for your approval before acting (US-only for now, on paid AI plans). Android is getting stricter too. From January 2027, developer verification rules go global, eventually covering apps installed from anywhere, not just the Play Store. Teams shipping outside Play should register now. New APIs automate it in CI. Details here.
Hijacked model answered Prime Intellect customers for two days: Yesterday, we covered the decentralized AI lab’s new Model Factory. Now it has a security mess on its hands. The lab says an attacker swapped its DeepSeek V4.1 Flash endpoint for a copy that refused just 1 in 100 jailbreak prompts. The swap ran November 8 to 10. Devs found out through OpenRouter and now want model attestation to verify you’re running the model you signed up for.

PRESENTED BY WIREMOCK
See how your agents behave against external services - without rate limits, flaky vendor sandboxes, or a dozen API keys.
• 130 simulated APIs, from Stripe and OpenAI to GitHub and Okta
• Hosted simulations based on real recorded API calls
• Free: no account, nothing to install, any key works

INSIGHT
A developer tried to replace Claude Code with a local setup. Here’s what broke

Source: The Code, Superhuman
Agent Memory. AI coding agents are great until they get stuck on the same bug, pile on broken code, and forget what they were building. Smaller local context windows even make it worse. Veteran engineer David Gewirtz wanted to see if a free, local setup could replace Claude Code, so he ran an experiment. He paired OpenCode with Qwen’s coding model on a 128GB Mac Studio. No subscriptions or cloud involved, which sounds promising, except his agent had just 64,000 tokens of context, versus Claude Code’s million.
The Build. First came a PDF organizer built on a new macOS 27 feature the AI had never seen. It read one PDF but couldn’t handle a folder. Next, a tool to rename his 7,000+ screenshots. He spent four days fighting the file list, with failed fixes eating context until the agent lost track. He then deleted the broken code, opened a fresh session, and tried again. It worked.
Here are his rules worth stealing:
If your agent keeps failing at the same bug, stop piling on patches. Scrap the code and start fresh.
Give each feature, bug fix, or test its own conversation so old mistakes stay behind.
Have your agent save what it learns in a markdown file, then feed that file to a fresh session to code.
Pick the model yourself instead of letting the agent grab whichever one it finds first.
Keep shell commands behind manual approval. Your agent can delete files as easily as it creates them.
Hidden Costs. Gewirtz spent nearly two weeks on an app he reckons he could have hand-coded in a few evenings. Free local agents can handle small projects, but long debugging sessions break them. If you’re ready to try it for yourself, Ollama’s OpenCode guide sets up the same stack with one command.

IN THE KNOW
What’s trending on socials and headlines

Meme of the day.
Hidden Glue: An ex-Stripe engineer argues the recent outages at Microsoft 365, airlines, and banks share one root cause. Something critical is quietly draining out of engineering teams.
Field Training: Forward Deployed Engineer might be AI’s most hyped job title right now. Anthropic just opened a 9.5-hour path teaching you exactly what the job takes.
Robot Manifesto: MongoDB’s co-founder says physical AI will stall without open source. And he actually makes a pretty interesting case for it (3.2K likes).
IPO Reveal: Anthropic’s record IPO just minted thousands of employee millionaires overnight. Here are the exact numbers on what the average employee is worth (3.2M views).
Stroke of Genius: Devs are obsessed with a hand-painted short retelling of coding history. It ends on a point even AI skeptics can’t argue with (3M views).
Automate This: A SpaceXAI engineer shows how to put your life on autopilot with Grok Bot and takes automation to a whole new level (2.6M views).
Mistake Mining: This prompt makes Claude Code read 60 days of its own mistakes, then fix the repeat offenders permanently (1.1K likes).

TOP & TRENDING RESOURCES
Top Tutorial
How to ship a 21,000-line feature with coding agents: Senior dev Matt Pocock walks through his workflow from requirements and architecture to implementation and review.
Top Tool
CodeRabbit Verify: Test agent-written code in a real running app before it ships.
Trending Cookbook
How to orchestrate coding agents without making them worse (by Google Cloud): Learn five research-backed patterns for splitting work, verification, parallel agents, model councils, and specialist reviewers, plus when multi-agent setups actually beat a single agent.

IN CASE YOU MISSED IT
Our most-clicked story from yesterday
A dev recreated the entire planet in Minecraft at 1:1 scale, furnished interiors included. Type your real address and walk through your own neighborhood.
Whenever you're ready to dive deeper
If you enjoyed this newsletter, you should definitely check out our IT and cybersecurity newsletters. Subscribe using the links below:
We put together 20+ hands-on guides covering coding agents, agentic engineering, and leadership frameworks to help you level up your career. You can get them here.
Advertise with us: Join companies like Google, IBM, and Datadog. Showcase your product to our 350K+ engineers and 150K+ followers on socials. Get in touch.
What did you think of today's newsletter?
You can also reply directly to this email if you have suggestions, feedback, or questions.
Until next time — The Code team




