Meta Q2 revenue came in at $60.8B (up 28%), above the $60.2B consensus, but $31B in capex collapsed free cash flow 91% to $784M. EPS landed at $6.18 against a $7.17 estimate, with the miss driven by $2.4B in legal charges and $1.18B in severance. The Q3 guide of $61-64B undershot the $63.24B consensus, and that cut, not the topline beat, dropped the stock 9.6% after hours to $529.15, per Investing.com.

Why it matters. Same evening as Microsoft, the market rewarded spend that shows revenue and punished spend that does not. Read our take.

"Pacing the Frontier" published July 28 with 1,293 signatories from frontier labs, among them Anthropic CEO Dario Amodei, OpenAI chief scientist Jakub Pachocki, SSI CEO Ilya Sutskever, and DeepMind co-founder Shane Legg. Both Anthropic and OpenAI backed it at the company level, the first time sitting frontier-lab leadership has endorsed a government pacing mechanism. The ask is narrow: the US should support an international effort to build the technical and governance tools needed to deliberately pace automated AI research, per Fortune.

Why it matters. When the people building the systems want a coordinated brake before they need one, plan for a regulatory environment that tightens, not one that loosens. Read our take.

Article 50 of the EU AI Act binds all 27 member states on August 2. Two duties land at once: any conversational AI must disclose it is an AI before the first exchange, and any system generating synthetic audio, image, video, or text must embed machine-readable markers. Systems already deployed get a transitional window to December 2 for content marking. The Code of Practice signatory window closed July 27, so signatories get a presumption of conformity and non-signatories lose the safe harbor, per the EU AI Act service desk.

Why it matters. If you ship any chatbot or generative feature to EU users, the disclosure and watermarking clock runs out this weekend. Read our take.

OpenAI launched ChatGPT for Academic Researchers on July 29. Qualifying faculty and postdocs get GPT-5.6 Sol Pro, ChatGPT Work, and Codex at no cost through 2027, roughly $200 per month of value per seat. The first 10,000 seats go out this summer, with Princeton's Institute for Advanced Study and France's Ecole normale superieure already in. It is part of a $250M-plus commitment and does not include model weights or training access, per OpenAI.

Why it matters. Free frontier access for the people who write the training-data papers is a cheap way to own the research funnel; your next grad hires will arrive already fluent in it. Read our take.

Nvidia launched the Open Secure AI Alliance on July 27 with more than 50 founding partners, donating open-weight models and the NOOA agent harness framework, per Nvidia's announcement. Absent from the roster: OpenAI, Google, and Anthropic, despite OpenAI's evaluation model having triggered the Hugging Face breach that motivated the coalition. The exclusion of the three biggest US frontier labs reads as pointed given the concurrent Pacing the Frontier debate.

Why it matters. Security standards are becoming a turf line; if your stack leans on one of the absent labs, expect two competing "secure AI" checklists to reconcile. Read our take.

Point AI agents at Make to automate the manual reporting an earnings week generates, from filings to a formatted summary. Try Make.

AI meeting notes that capture an earnings or analyst call without dropping a bot into the room. Try Granola.

We compared Braintrust, LangSmith, Promptfoo, and Arize on CI gating, trace capture, self-host, and switch cost, so you pick on lock-in rather than logo. For any team standardizing eval before it scales. Read the comparison.

The decision turns on durable-execution substrate, not token price. If you need replayable long-running agents, that narrows the field fast. For platform leads picking an agent framework. Read the guide.

Go from a single duration metric to span-level agent traces using the OpenTelemetry GenAI semantic conventions. For platform teams instrumenting agents in production. Read the guide.

A vendor-vetting process, scope policy, and monitoring checklist for enterprise MCP. For the person who has to approve the tool server. Read the guide.

What actually isolates an agent's blast radius when a tool call returns hostile output. For anyone shipping agents with real credentials. Read the guide.