Jason Says
Today's biggest signal: OpenAI and Anthropic are both racing toward IPO. The era of AI 'money-burning black holes' is ending — the next phase is about who can actually generate revenue. For indie developers in the AI space, this is both pressure and a window of opportunity.
OpenAI has confidentially submitted its S-1 filing to the SEC, just over a week after Anthropic did the same. With both AI giants sprinting toward public markets, the industry is formally transitioning from burn-rate competition to public accountability. Valuations and monetization strategies will face open scrutiny for the first time.
Apple's WWDC 2026 headline moves include waiving cloud AI API fees for developers whose apps have fewer than two million downloads, plus major AI upgrades to Siri, Safari, Shortcuts, and Photos. The low-cost strategy clearly targets indie developers, positioning Apple to leapfrog competitors in building out its AI app ecosystem.
Anthropic announced that Claude can now be embedded within Apple's Foundation Models framework, letting developers natively invoke Claude to build intelligent apps on Apple platforms. This marks Claude's expansion beyond web and API access into on-device AI, deeply integrating with the Apple Intelligence ecosystem.
The open-source skill '/last30days' enables AI agents to simultaneously search Reddit, X, YouTube, Hacker News, Polymarket, and the broader web, ranking and summarizing results by real user engagement — likes, shares, and actual money. Installable in one click via Claude Code's Marketplace with auto-updates, it's one of the most comprehensive trend-research agent skills available.
Nous Research launched Hermes Agent with a built-in learning loop — automatically creating and refining skills after each conversation, searching memory across sessions, and progressively building a user profile. It runs on a $5/month VPS or serverless environment, representing the most complete open-source implementation of continuous self-improvement in an agent to date.
CORE (Contrastive Reflective algorithm) trains models to compare their own past reasoning steps, identify right-versus-wrong differences, and self-correct rapidly — without the large-scale training required by RLVR. Significant reasoning gains are achievable with very few samples, potentially slashing fine-tuning costs and enabling strong reasoning models from small datasets.
Researchers introduced Imagined Perception Tokens (IPT), allowing vision-language models to generate intermediate representations of virtual viewpoints during inference. This tackles occlusion and cross-angle spatial reasoning challenges. Developers building robot navigation, AR scene understanding, or 3D applications could see direct spatial capability gains without swapping out their base models.
研究分析了 35,361 条明确标注 AI 使用的 GitHub 代码注释，构建出首个「AI 辅助开发行为分类体系」，揭示开发者如何修改、适配 AI 生成代码。结论是：AI 输出普遍被二次改造，而非直接使用——为 Copilot/Claude Code 等工具的下一步优化方向提供了实证基础。
Subscribe for daily AI updates + free playbook
📘 Subscribe Free