Disney Is Reportedly Buying Pieces of Character AI for $20M

1  Disney moves to pick up Character AI's staff and assets after its bankruptcy

Five years after launch, chatbot platform Character AI filed for bankruptcy, pointing to high costs and legal fights. Disney is reportedly in talks to buy some of its assets for $20M and hire most of its team.

2  The FBI says the accused OpenAI spy kept a folder called "Saudi gov docs"

New court filings in the case against the former OpenAI employee accused of bugging an executive conference room reportedly show contact with the Saudi royal family and plans to move back to the country. The case is still under investigation. In other OpenAI news, ChatGPT Slides now supports real-time collaboration and direct editing on Pro, Business, and Enterprise plans. Try a few prompts.

3  A hack on OpenCode exposes developers' AI provider keys

Attackers spent about 48 hours inside an old OpenCode system and took the provider keys of 1.4% of its Console users. Some stolen OpenAI and Anthropic keys were drained, with bills reported as high as $1,700, and OpenAI is refunding credits. If you use OpenCode, here is what to do now: rotate your keys and check your usage.

4  The US bans DeepSeek from federal agencies as its cheap coding model gains ground

Homeland Security barred DeepSeek from federal civilian agencies, part of a wider push to keep Chinese technology out of critical systems. The same week, DeepSeek released V4.1-Flash-1017, a budget model that its own tests say slightly beats Claude Opus 5 on a coding benchmark, at $0.30 per million input tokens. See the API docs.

5  China's economy grows 4.5% in the third quarter

China's economy grew a little faster than expected, helped by strong exports even as spending at home stayed weak.

What hiring managers look for. Google VP Josh Woodward reportedly looks for people who learned something new and changed course because of it. Proof that you can adapt now counts for more than a long list of past roles.

Advice from people who just switched jobs. Anthony Maggio, who moved from Airtable to become chief product officer at Runway, built a personal app to show his hands-on skills instead of describing them. Chandra Raghavan, who went from Meta to Google, warns against a view of AI "stuck on ChatGPT in 2022." And Gadi Ben-Zvi of Eli Lilly's AI division says what you can do today matters more than your credentials.

The questions you will face. Interviewers increasingly ask where an AI project failed, what was harder than expected, where AI was overused, and what you would do differently. Even job search advice now goes out of date within months.

The squeeze at the bottom. Anthropic's own hiring report shows junior engineers' share of technical hires fell from 27% to 9% since early 2025. One in five employees has moved from doing the work to managing Claude, and technical staff now spend 16% more time in meetings. The open question is how people starting out will build experience when the entry-level work is the first thing AI takes over.

Meta: no proof, no merge. A small Meta team used an AI agent to build Forum, a standalone Groups app, to feature parity with Facebook Groups in eight weeks. Their rule, explained in Meta's write-up: a task is done only after automated tests and visual checks pass, and every pull request includes a test plan and a recording that proves it works. Stress tests caught a memory problem that the agent then fixed, cutting memory use roughly in half. Engineers still approve every change, because passing tests does not guarantee good code.

Make agents wait for the tests. Claude Code's hooks guide shows how to stop an agent from marking work finished until the tests pass. Anthropic also explains how it handles code review: the agent checks its own work first, a fresh AI reviews it, and human approval scales with the risk.

Spend effort where it pays off. SpaceX AI's Lauren (poteto) argues for an upside-down pyramid: put the most work into a codebase that makes mistakes hard to write, then lint rules and tests, then written rules for agents, and rely least on human review, which cannot keep pace.

Replace line-by-line review. Addy Osmani makes a similar case (2.6K bookmarks): when agents write code faster than people can read it, use end-to-end tests, property tests, and AI reviewers that challenge each other, and let humans focus on intent and risk.

🍿 Pitch Yourself. Designer Darshita Patankar made a short film to promote herself to employers (20K likes), a fitting example of today's career advice in action.

🤔 More Code, Same Profits. AI researcher François Chollet lists four observations: there are more software engineers than in 2025, they work harder than ever, agents now write almost all the code, and most companies see no change in profits yet (2.7K likes).

🎮 No Going Back. Epic CEO Tim Sweeney says the company is moving developers onto AI-assisted work to ship more, not to shrink the team (6.2M views).

🌊 Built With Claude. Anthropic highlighted six recent projects, led by a wildfire tracker, while one Redditor built a realistic ocean simulator in a 61 KB file (2M views).

🧸 Learning Through Play. A new Claude skill turns everyday objects into hand-drawn, interactive lessons for kids, starting with a washing machine (5.7K bookmarks).

💾 Boxed AI. A Reddit concept imagines Anthropic and OpenAI sold as 1985 software boxes on a computer store shelf.

📝 House Rules for AI. Former Amazon engineer Steve Huynh breaks down the internal AI rules one engineering lead set for his team.