Sponsored by

OpenAI's first gadget is the $230 Codex Micro macropad

Key Takeaway: OpenAI has partnered with Work Louder to launch the Codex Micro, a customizable mechanical macropad designed to help developers monitor and control multiple Codex coding agents.

More Insights:

• Six illuminated “agent keys” show whether agents are idle, thinking, finished, or experiencing errors.

• Users can select agents, approve or reject code changes, branch tasks, and activate voice controls.

• The device supports USB-C and Bluetooth, with silent or clicky mechanical switches.

• Nearly every key, dial, joystick, light, and software layer can be customized for Codex or other apps.

• At $230, it is largely a redesigned version of Work Louder’s existing macropad rather than OpenAI’s anticipated Jony Ive hardware project.

Why it matters: The Codex Micro suggests that AI agents are becoming active coworkers that need dedicated controls and status indicators—but its real significance depends on whether physical interfaces improve productivity or simply turn AI software into expensive desk accessories.

Know whether your secrets setup is ready for AI agents

AI agents are minting credentials faster than teams can track them, and secrets are already scattered across Vault, cloud managers, CI/CD, and .env files nobody rotates.

This webinar will show you how to:

• 🔍 Surface the shadow secrets hiding in your environment

• 🤖 Stand up machine identities and workload auth built for agents

• ✅ Pressure-test your setup before agents hit prod

Walk out knowing exactly what modern teams are doing to build a more developer-friendly credential control plane.

Chinese AI model takes US tech industry by surprise with abilities rivaling Claude and ChatGPT

Key Takeaway: China’s Moonshot AI has released Kimi K3, a powerful open-weight model that appears competitive with leading U.S. systems while raising fears that cheaper Chinese AI could weaken America’s technological and financial advantage.

More Insights:

• Kimi K3 reportedly rivals top models from OpenAI and Anthropic, particularly in coding.

• Semiconductor stocks fell sharply as investors questioned whether massive AI infrastructure spending will deliver adequate returns.

• The release revived memories of DeepSeek’s 2025 debut, which triggered a major sell-off in U.S. technology stocks.

• Chinese firms are advancing despite U.S. chip restrictions by improving efficiency and developing domestic hardware alternatives.

• Kimi K3 could pressure the pricing and revenue models of closed AI providers, even if the market reaction proves exaggerated.

Why it matters: Kimi K3 suggests that leadership in AI may depend less on who owns the most expensive chips and more on who can build capable, affordable, and widely accessible models—potentially transforming today’s AI arms race into a global competition over efficiency, openness, and adoption.

Former OpenAI CTO builds open weight model in 9 months

Key Takeaway: Thinking Machines Lab’s first model, Inkling, challenges closed, one-size-fits-all AI by giving organizations a powerful open-weight foundation they can customize with their own expertise.

More Insights:

• Inkling has 975 billion total parameters, with 41 billion activated per task for greater efficiency.

• The multimodal model was trained on 45 trillion tokens and can process text, images, audio, and video.

• It performs strongly in coding, tool use, safety, and agentic tasks, but still trails leading Chinese models on several benchmarks.

• Companies can download and fine-tune Inkling, while Thinking Machines plans to earn revenue through its Tinker customization platform and hosting ecosystem.

• Early results suggest specialized models can outperform proprietary rivals at a fraction of the operating cost, though the company’s claims have not been independently verified.

Why it matters: Inkling represents a bigger shift than another benchmark competition: if businesses can turn open models into private, specialized systems that are cheaper and better at their own work, the future of AI may belong not to the company with the single smartest model, but to the ecosystem that makes intelligence easiest to customize.

Other stuff

• I interned at OpenAI in San Francisco. Here's how I landed it, what it was like, and what I thought of Sam Altman.

• I Gave an AI Agent Access to My Passwords. Here's What Happened.

• Anthropic's newest ad is creeping people out

• GPT-5.6 used a prompt to close a 30-year gap in convex optimization

• Over 10k people shared what they love about GPT-5.6

• I Unlocked ChatGPT's Dark Side for $101.40 [video]

• The CIA Operative Who Spied on the U.A.E.–and Played a Role in Its AI Win

• For Software Engineers, the AI Reckoning Is Here

Hope you enjoyed today's newsletter

⚡️ Join over 300,000 people using the Superpower ChatGPT extension on Chrome and Firefox.