There is a Reddit thread making the rounds that compares agent frameworks. It covers more than twenty of them. LangChain, LangGraph, LlamaIndex, Haystack, Semantic Kernel, Pydantic AI, DSPy, Mastra, AutoGen, CrewAI, MetaGPT, the OpenAI Agents SDK, Google ADK, Strands Agents, CAMEL, AutoGPT, AgentScope, DeerFlow, and the list keeps scrolling. Every
Somewhere in the last eighteen months, 'AI-first' completed the full hype-word life cycle. Fresh, then fashionable, then mandatory, then meaningless. Record time. An investor at Sopra Steria's venture arm put it plainly: they now have to differentiate between 'a SaaS company augmented by
In late July, Nvidia launched the Open Secure AI Alliance with Microsoft, SpaceX, IBM, and Palantir as founding members. Within days the group swelled past 30 companies, pulling in Adobe, BlackRock, Cisco, Intel, and Visa. One week later, TechCrunch ran this headline: 'Nvidia doesn't mess around: A
AI took nearly half of all global startup funding in 2025. That is $202.3 billion out of roughly $425 billion, up from 34 percent in 2024. Q1 2026 pushed the AI share to something like 80 percent of new capital. The OECD, being the OECD, framed 2025 more conservatively
Nine open-weight models shipped in twelve days this July. The headliner was Thinking Machines' Inkling, a 975 billion parameter model released with weights on day one, from a lab that could have charged rent on a closed API and chose not to. Five different vendors put frontier or
Every founder I meet wants the glamorous bet. Build the next lab. Raise the megaround. Chase the frontier. I want to make the case for the opposite, the bet nobody brags about at dinner: build your product on open weights and let the richest companies in history pay for your
On July 27, Moonshot released the full weights for Kimi K3. All 2.8 trillion parameters, 104 billion active. The benchmarks are real, the model is a monster, and the download is free. What it is not, despite what half the coverage says, is open source. K3 shipped under a
Let me be clear about where I stand before I say anything else. I build on open models. I want them to win. The strongest argument for open AI is that it lets people like us build companies without asking a lab for permission. This piece is not against open
On August 2, 2026, days from now, the EU AI Act's transparency provisions take effect. That date matters even if your company has never shipped anything to Europe, because the companies you sell to have, and their compliance requirements flow downhill to you. The obligations are specific: documented
On July 16, Moonshot launched Kimi K3, a 2.8 trillion parameter model and the largest open model ever released. Demand was so intense that Moonshot's first-party API sold out within 48 hours. Then this weekend the full open weights went live, all 594 gigabytes of them,
On July 24, Jensen Huang did something he had never done before. The CEO of the world's most valuable company made his first post on X. It was not a product announcement or a victory lap. It was a three page policy letter to Washington titled Open Weights
On July 24, Anthropic shipped Claude Opus 5. It lands within 0.5% of Fable 5's peak intelligence at half the cost per task, priced at $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8. It set a new state
On July 22, the White House got into the business of model training pipelines. OSTP chief Kratsios publicly accused Moonshot AI of distilling Anthropic's Fable model without authorization to build Kimi K3. Treasury Secretary Scott Bessent said sanctions remain on the table. The Hill added a second charge,
Every pitch deck I've seen this year has the word autonomous in it. Usually in the first three slides, usually in a font size that suggests the founder believes it. The agentic AI market is projected to grow from $7.8 billion to $52 billion by 2030. Gartner
On July 16, Moonshot released Kimi K3, a 2.8 trillion parameter open-weight model, the largest ever put out in the open, benchmarked directly against the top closed systems in the US. Forty-eight hours later, Moonshot paused new subscriptions because demand had pushed their GPU capacity to its
The self-driving company is the pitch of the summer. Forbes profiled it on July 20 as AI's next business model, pointing to Replit, where agents investigate production incidents, answer product questions, analyze business data, enrich sales leads, build presentations, and triage support tickets. MIT Sloan says these
Everyone in AI is staring at the wrong chart. The benchmark leaderboards get the headlines, the model releases get the launch videos, and meanwhile the chart that actually determines which startups survive is quietly going vertical. Inference cost for GPT-4-class capability fell from $30 per million tokens in
Last week, on July 15, Thinking Machines Lab shipped its first model. Mira Murati's lab called it Inkling, and on paper it looks like another big release: 975 billion parameters in a Mixture-of-Experts design, 41 billion active per token, multimodal, Apache 2.0 open weights. But
Last week settled an argument I have been having with people for two years. On July 15, Thinking Machines Lab, Mira Murati's outfit staffed heavily with ex-OpenAI leadership, shipped its first model. Not a demo, not a waitlist, not an API with usage tiers. Inkling landed as
Let me state the tradeoff honestly, because most people arguing my side won't. Open-source models lag the closed frontier. The gap is real, roughly three to six months depending on the task. If you rent frontier capability from a closed API, you get the best model on
There is a slide deck making the rounds in 2026, and if you have sat through one enterprise AI pitch this year, you have seen it. Map your workflows. Deploy agents to automate them. Collect the savings. Every consultancy has a version. Every agent platform has a version. The logos
On May 19, Andrej Karpathy joined Anthropic. He co-founded OpenAI, ran AI at Tesla, and is probably the most recognizable AI researcher alive who isn't running a lab. His own explanation was simple: "the next few years at the frontier of LLMs will be especially formative.
Somewhere in 2026, 'AI-first' became the default startup identity. It's on every landing page, every pitch deck, every hiring post. And I think it's quietly hurting the founders who lean on it hardest, because it describes the ingredient, not the business. Nobody in
On April 29, the House Committee on Homeland Security and the House Select Committee on China sent letters to Anysphere, the company behind Cursor, and to Airbnb. The committees want to know why American companies are running Chinese-developed AI models like Alibaba's Qwen and Moonshot's