Daily AI RoundupsThursday, 16 July 2026

The 16 Jul 2026 edition leads with Thinking Machines Lab releasing Inkling, a 975B-parameter open-weight multimodal model under Apache 2.0, and TSMC planning another $100B in US chipmaking after a 77% profit jump on AI demand. It also covers Apple Intelligence clearing China launch with Alibaba's Qwen, Variant's model-foundry thesis, Simon Willison's read on xAI open-sourcing Grok Build after repository-upload backlash, Anthropic's investor meetings after confidentially filing for an IPO, OpenAI's $230 Codex control pad, Anthropic-backed Ode's forward-deployed engineering model, Ring-Zero's 1T-parameter zero-RL paper, and agent-workflow arguments from Boris Cherny, Charlie O'Neill and Cursor's Lee Robinson.

• Open models
• AI chips
• China AI
• AI infrastructure
• AI IPOs
• Coding agents
• Zero RL
• Agent workflows

Stories in this roundup

1. 01
   Mira Murati's Thinking Machines Lab releases Inkling, a 975B-parameter open-weight multimodal model under Apache 2.0TechCrunch
2. 02
   TSMC to invest another $100 billion in US chipmaking as Q2 profit soars 77% on AI demandCNBC
3. 03
   Apple Intelligence approved for launch in China, with Alibaba's Qwen AI set to be integratedTechCrunch
4. 04
   Variant's Elijah Fox and Daniel Barabander argue the next AI infrastructure fight is dedicated model foundries, not one-size-fits-all model APIsVariant/X
5. 05
   Simon Willison analyzes xAI's open-sourced Grok Build codebase after the CLI was found uploading full repositories and commit historySimon Willison
6. 06
   Anthropic lines up investor meetings ahead of a potential October IPO, after confidentially filing its IPO prospectus with the SEC last monthCNBC
7. 07
   OpenAI ships its first hardware product, a $230 control pad for the Codex coding agentThe Verge
8. 08
   Anthropic-backed Ode launches, betting the next trillion-dollar AI business is embedding forward-deployed engineers inside enterprisesTechCrunch
9. 09
   Ring-Zero scales zero-RL reasoning to a 1T-parameter model and reports emergent behaviors including self-verification, parallel reasoning and context anxietyarXiv
10. 10
    Boris Cherny says teams should turn domain knowledge into CLAUDE.md files, skills, docs and automation so agents can work in codebases with near-zero extra contextBoris Cherny/X
11. 11
    Baseten's Charlie O'Neill argues continual learning means moving agent knowledge out of short-lived context windows and into models that improve on the jobCharlie O'Neill/X
12. 12
    xAI sues a user for generating CSAM deepfakes with Grok, marking a rare platform enforcement action against AI-generated abuseThe Verge
13. 13
    Cursor's Lee Robinson explains recursive model improvement, from user-feedback loops and private evals to models helping train the next modelAI Engineer