← Back to News
DAILY DIGEST

Week In Review June 22-27, 2026 - The Week AI Got Embedded In Industries It Used To Threaten

Google DeepMind bought into A24, OpenAI declared advertising a core business, Gemini shipped Deep Think, the Transformer's co-author landed at OpenAI, and a new agent attack class hit thousands of orgs. Five days that all pointed the same direction - AI moving from demo to deployment inside real industries.

By Michael Eakins•• min read
Weekly DigestGoogle DeepMindA24OpenAIGeminiAI Security

The Thread That Tied Everything Together

The week of June 22 through 27, 2026 will read, in retrospect, as the week the AI story stopped being about model capability and started being about embedding. Almost every major item that landed was a frontier lab pushing deeper into a real industry or workflow rather than announcing a benchmark. Google DeepMind put money into a film studio. OpenAI declared advertising a core business and shipped audio agents. The Transformer's own co-author took a job redesigning it. And a new attack class showed exactly how fragile the deployment layer still is.

This digest covers the week in review, then previews what to watch in the week ahead.

Week In Review

1. Google DeepMind Invested ~$75M In A24 To Build Filmmaking Tools

The marquee story of the week. Google invested a reported $75 million in the independent studio A24 and paired it with a Google DeepMind research partnership to build AI tools for filmmakers — storyboards, previsualization, and production workflow. Crucially, the deal does not give Google access to A24's content library or data, the tools are explicitly not a system for generating finished films, and A24's filmmakers retain creative control. Three years after the strikes, it is the clearest picture yet of how AI enters creative production without re-triggering the replacement fight.

For the full analysis of why the terms — not the dollar figure — are the template the rest of the industry will copy, see my deep dive on Hollywood's AI détente.

2. OpenAI Declared Advertising A Core Business

OpenAI confirmed that advertising is now a core part of its strategy, with ChatGPT serving more than nine hundred million weekly active users and roughly one in five queries carrying direct commercial intent. The company has been building toward this since opening its self-serve ChatGPT Ads Manager to U.S. businesses earlier in 2026, and this week's framing made the strategic intent explicit: the answer engine is becoming an ad engine. Alongside it, OpenAI launched three real-time audio models for conversational agents, translation, and transcription — a push to make voice agents production-ready.

3. Google Shipped Gemini 2.5 Pro With Deep Think

Google launched Gemini 2.5 Pro with a Deep Think reasoning mode, positioned as its most capable model to date. The release lands amid a broader pattern of Google trying to make Gemini the operating layer for daily tasks — a proactive app, managed agents in the Gemini API, and expanded creation workflows in AI Studio. The model news matters less than the platform direction: Google wants Gemini to be where work happens, not just a model you call.

4. The Transformer's Co-Author Joined OpenAI

Noam Shazeer, one of the eight authors of the 2017 "Attention Is All You Need" paper, joined OpenAI as Lead for Architecture Research — a role explicitly aimed at what comes after the Transformer. Days earlier, AlphaFold lead and Nobel laureate John Jumper left Google DeepMind for Anthropic. Two foundational researchers leaving the same lab in the same window moved Alphabet's stock and crystallized a shift from a compute war to an architecture war.

I unpacked what this means for anyone building on top of these models in my analysis of the architecture reset.

5. ChatGPT Fell Below Half The Assistant Market

For the first time, ChatGPT's share of the global AI-assistant market dropped under fifty percent, to roughly 46.4 percent, while Google Gemini rose to about 27.7 percent and Claude reached around 10.3 percent. The more interesting number sat underneath: Claude reportedly converts about thirteen percent of users to paid subscriptions, the highest paid-conversion rate in the field. Breadth is consolidating toward Google while depth and monetization concentrate elsewhere — the market is segmenting, not crowning a single winner.

6. "Agentjacking" Hit Thousands Of Organizations

A new attack class disclosed this week, dubbed agentjacking, reportedly achieved an eighty-five percent exploitation rate and affected more than two thousand three hundred organizations. The technique plants markdown-injected instructions inside fake error reports that AI coding agents read as legitimate debugging guidance and then execute. It is a sharp reminder that the deployment layer — agents acting with real permissions — remains the soft underbelly of the agentic era, and that prompt-injection-style threats have graduated from theory to mass exploitation.

7. Thomson Reuters Warned Of The Cost Of Not Adopting

Thomson Reuters released its 2026 Future of Professionals report, warning that failing to implement AI effectively puts as much as one hundred forty-three billion dollars in U.S. client revenue at risk. The framing is a notable inversion of the 2023 anxiety: the report treats the larger danger as moving too slowly, not too fast. Whether that is a genuine signal or a vendor-shaped narrative, it captures a real shift in how professional-services firms are now talking about AI internally.

8. Apple Detailed iOS 27 Apple Intelligence Features

Apple continued filling in the practical edges of Apple Intelligence in iOS 27 — AI-assisted bill splitting, automatic password replacement after breaches, calendar creation from natural language, smarter Shortcuts, and AI-organized Safari tabs. None of it is frontier, and that is the point: Apple is competing on quietly useful on-device features rather than raw model leadership, the same posture it signaled at WWDC earlier in the month.

The Pattern Underneath The Week

Read the eight stories together and the through-line is embedding, not capability. The A24 deal embeds a lab in film. OpenAI's ad business embeds ChatGPT in commerce. Gemini's platform push embeds Google in daily work. Even the architecture-talent moves and the agentjacking disclosure are about the same underlying transition — from models you admire to systems you deploy, with all the strategic value and operational risk that shift carries. The frontier is no longer mostly about who has the smartest model. It is about who is woven most deeply into the industries and workflows where the model actually gets used.

What To Watch In The Week Ahead

  • A second studio-lab deal. The A24 template is now public. Watch whether a larger studio signs a similar investment-plus-data-wall arrangement — the signal that a market, not a one-off, is forming.
  • GPT-5.6. OpenAI previewed it for late June. Whether it ships on schedule and how it is positioned against Gemini 2.5 Pro Deep Think will set the tone for the summer model cycle.
  • Agentjacking fallout. Expect rapid patching guidance and vendor responses; watch whether coding-agent providers ship input-sanitization defaults or treat it as a user problem.
  • Ad-load signals in ChatGPT. With advertising now declared core, watch for the first visible changes in how commercial answers are presented to OpenAI's nine hundred million weekly users.