The AI Brief

Alphabet reported Q2 2026 revenue of $119.8 billion, up 24% year-over-year and above analyst consensus of roughly $116.9 billion. Operating income rose 30% to $40.8 billion, driven by AI-led growth across Search, YouTube, and Cloud, but the company raised full-year capex guidance to $195 billion–$205 billion and flagged continued supply constraints and margin pressure ahead. Alphabet's stock fell roughly 4% in after-hours trading on the guidance raise.

Cloud revenue surged 82% year-over-year to $24.8 billion, with backlog reaching $514 billion. Q2 capex totaled $44.9 billion, primarily in technical infrastructure, and free cash flow turned negative at $5.9 billion for the quarter. Approximately 60% of the Q2 capex went to servers, with 40% to data centers and networking equipment.

The company raised full-year 2026 capital expenditure guidance to $195 billion–$205 billion from $180 billion–$190 billion. CFO Anat Ashkenazi told analysts: "The increase in the range is primarily due to an acceleration in the delivery of capacity to meet growing demand." Pichai noted that reaching Gemini 4, described as the next "big leap forward," will require "a massive leap in base model scale." Nearly 90% of the Fortune 100 is now using Gemini Enterprise.

SoftBank secured robust support for its $40 billion bridge loan, with 21 new institutions committing $7 billion despite concentration risks regarding its OpenAI investments. The bridge loan was raised to fund SoftBank's direct investment in OpenAI and is the largest AI-specific debt instrument yet assembled in global credit markets.

The syndication adds institutional depth to a structure that analysts have flagged as circularly collateralized: SoftBank's primary asset backing the loan is the OpenAI equity position the loan was used to acquire. The 21 new lenders join an original bank group, bringing total institutional participation well above two dozen counterparties. Concentration risk disclosures in the syndication materials explicitly named the OpenAI position as the dominant exposure.

On July 21, 2026, Anthropic shipped "Record a Skill" inside Claude Cowork, a feature that lets Pro, Max, and Team users screen-record themselves doing a task once, then has Claude analyze that recording and turn it into a reusable, rerunnable skill. No prompt engineering, no manually describing the steps, you just do the thing, and Claude figures out how to automate it.

Claude's underlying system captures screen activity, mouse clicks, keystrokes, and voice commentary, then processes this demonstration into a structured, reusable skill definition. Once captured, the skill appears in the user's skill library and can be triggered again without repeating the manual walkthrough. Skills are single repeatable procedures; plugins are full roles bundled with tools, Record a Skill produces only the former.

Anthropic's Record a Skill turns Claude Cowork into an on-ramp for non-technical workers to build automation without writing prompts or connecting APIs, and the Team-plan gate makes recorded skills shareable across organizations, placing Anthropic in direct competition with Zapier's workflow-library model. Privacy documentation specific to skill recordings had not been published as of the launch date; treat any recording session containing passwords, client PII, or regulated data as potentially at risk and keep it off-screen during recording.

Disclosure: Claude, which generates this brief, is built by Anthropic.

First covered in Vol. I, No. 61. Anthropic released Claude Opus 5 on July 24, 2026 at $5 and $25 per million input and output tokens, the same price as Opus 4.8 and half what Fable 5 costs. The model reaches roughly Claude Fable 5–level intelligence at half the price, adds a low/medium/high effort toggle so you can trade cost for capability per request, and sets new state-of-the-art scores on agentic-coding and knowledge-work benchmarks.

Internal benchmarks show it outperforms Fable 5 on several coding and knowledge-work tests, including a 43.3% score on Frontier Bench v0.1 versus Fable 5's 33.7%. On OSWorld 2.0 it surpasses Fable 5's best result at a third of the cost. The model features less restrictive cybersecurity safeguards than Fable 5, is not subject to the 30-day data retention policy, and is described as Anthropic's most aligned model to date.

Anthropic shipped its fourth model in under two months on July 24, 2026, after Mythos 5, Fable 5, and Sonnet 5 all landed in June, Opus 5 arrives as the model most teams will actually run every day. Fast mode doubles the rate to $10/$50 per million tokens and runs about 2.5× faster. Benchmarks carry a vendor-interest caveat: Anthropic ran its own Frontier-Bench harness using a GKE backend and Opus 4.8 as fallback on safety-classifier refusals, meaning headline scores are not pure single-model results.

Disclosure: Claude, which generates this brief, is built by Anthropic.

First covered in Vol. I, No. 63. On June 16, 2026, the European Parliament approved amendments that push most high-risk AI obligations out to December 2027 and August 2028, what many teams missed: August 2, 2026 still lands. Article 50 transparency duties, chatbot disclosure, AI-content marking, deepfake labeling, were not delayed, and the Commission's GPAI enforcement powers activate the same day.

Organizations that read the headlines about "deadlines being delayed" and quietly stood down their AI Act programs have made a serious mistake. The single most broadly applicable obligation in the entire Act, the Article 50 transparency duties, is not deferred. Neither is the enforcement machinery behind it. As of August 2, 2026, the AI Office and national authorities acquire the power to impose fines of up to €15 million or 3% of total worldwide annual turnover, whichever is higher, for breaches including Article 50.

What activates on August 2, per the Commission's guidance and multiple legal trackers, is the Commission's active enforcement toolkit, information requests, model access, and recall powers. The only carve-out: pre-existing systems get until December 2, 2026 on the machine-readable watermarking duty. Article 50(1) chatbot disclosure and Article 50(2) AI-content marking have no such grace period.