In partnership with
In Today’s Issue:
🤑 Google I/O turns Gemini into an agent stack
💵 KPMG brings Claude into enterprise delivery
📈 SynthID becomes a wider AI watermarking layer
📉 AI data centers hit local opposition
✨ And more AI goodness…
⚡ The Signal
Google I/O is turning the AI race from a model contest into a full-stack agent contest.
Google’s new I/O slate is not just another list of model names. Gemini 3.5 Flash is built around fast agentic execution, Gemini Omni pushes creation and editing across media, and Antigravity is becoming the developer layer where those models actually do work. That same story shows up outside Google, too. KPMG is embedding Claude into client delivery, SynthID is becoming shared provenance infrastructure, and local opposition to AI data centers is turning compute buildout into a political problem.
The frontier is no longer only who can make the smartest model. It is who can connect models, tools, trust, and physical infrastructure without the system buckling.
All the best,
Kim Isenberg
🏢 KPMG Puts Claude Inside Client Delivery
KPMG and Anthropic announced a global alliance that embeds Claude Cowork and Managed Agents inside KPMG Digital Gateway, starting with tax, legal, and private-equity work. The scale is the point. KPMG says its 276,000-person global workforce will get access to Claude, while the firms co-develop Claude-powered products for portfolio companies.
👉 tl;dr: Claude is moving from chat window to client-delivery infrastructure, where auditability, governance, and domain workflow matter as much as model quality.
🧾 SynthID Becomes Shared AI Provenance Plumbing
Google’s SynthID watermarking technology is being adopted by OpenAI, Nvidia, and others as AI-generated media gets harder to distinguish by sight alone. The important shift is institutional. Provenance is moving from one lab’s trust feature into a wider ecosystem layer for media, model outputs, and platform accountability.
👉 tl;dr: The AI industry is quietly standardizing around invisible provenance because visible disclosure cannot keep up with generated media.
🏗️ AI Data Centers Hit the Local Backlash Wall
The updated data-center story points to the infrastructure problem under every AI launch. People may like AI tools, but many do not want the compute facilities near them. Gallup’s latest survey found seven in 10 Americans oppose building AI data centers in their local area, with concerns around water, power, bills, pollution, and quality of life.
👉 tl;dr: AI’s next bottleneck may not be model talent. It may be whether communities accept the physical footprint required to run the models.
Ask your AI assistant to turn any big product-launch recap into three buckets: model capability, product surface, and infrastructure dependency.
Why it helps: Google I/O is a perfect example of the problem. If you only track the model names, you miss the stack around them: developer tools, provenance, commerce, Search, video, and the data centers underneath.
Try this: Paste a launch post and ask: "Separate the announcements into model upgrades, user-facing products, developer/infrastructure changes, and trust/safety mechanisms. For each bucket, explain what changes for a startup building on this ecosystem."
🎬 Watch This
Google I/O '26 Keynote is the official Google keynote for this year’s developer conference. The video is the right match for today’s issue because it anchors the Gemini 3.5 Flash, Gemini Omni, Antigravity, Search, creative tooling, and agentic product announcements in the company’s own launch narrative.
The week’s grittiest AI dustup is not a chatbot behaving badly. VentureBeat reports that four AI supply-chain incidents in 50 days hit OpenAI, Anthropic, and Meta-adjacent infrastructure, exposing release pipelines, dependency hooks, CI runners, and package-publish gates that model red teams usually do not test. The uncomfortable lesson is that AI security cannot stop at jailbreaks and system cards; the software factory around the model is now part of the attack surface.
Google’s I/O Stack Is Built for Agents,
Not Just Answers
The Takeaway
👉 Google announced Gemini Omni and Gemini 3.5, positioning Omni around multimodal creation and 3.5 Flash around agentic workflows.
👉 Antigravity is becoming the practical developer layer: the place where models move from helping write code to helping execute work.
👉 The same I/O collection points to agents across Search, the Gemini app, shopping, Workspace, YouTube, Chrome, and science tools.
👉 The fresh angle is not "Google has another model." It is that Google is trying to connect models, products, tooling, provenance, and commerce into one agentic platform.
Google I/O 2026 reads like a map of where the AI race is going. Google announced Gemini Omni for richer multimodal creation and editing, plus Gemini 3.5 Flash as the first model in a new family focused on frontier intelligence with action. That framing is important because it moves the story beyond a single chatbot upgrade. Google wants models that can sit inside workflows and do things repeatedly.
Antigravity is the clearest signal. Google describes the platform as moving beyond tools that help developers write toward agents that help them act.
The surrounding announcements make the same point across the rest of the company: information agents in Search, proactive help in the Gemini app, Universal Cart for shopping, Google Pics for Workspace, Ask YouTube, Chrome updates, and Gemini for Science all turn the model into an operating layer.
That is also why this should not be framed as a repeat of yesterday’s Flash economics story.
The new question is architectural: can Google make agents feel native across the surfaces people already use, while also solving trust signals, developer routing, media provenance, and cost? If it can, I/O 2026 is less a product launch than a platform reset.
Why it matters: The next platform war will be won by the company that makes AI useful inside daily workflows, not just impressive in demos. Google’s advantage is that it can wire agents into Search, Android, Workspace, YouTube, Chrome, and developer tools at once.
Sources:
🔗 https://blog.google/innovation-and-ai/technology/developers-tools/google-io-2026-collection/
🔗 https://www.youtube.com/watch?v=wYSncx9zLIU
🔗 https://arstechnica.com/google/2026/05/google-announces-agent-optimized-gemini-3-5-flash-and-a-do-anything-model-called-omni/
Have you been burned by a confident-sounding AI answer yet?
The chart: Artificial Analysis ranks Gemini 3.5 Flash below the absolute top frontier models in raw intelligence, but the second chart shows why it matters: it sits in the most attractive quadrant, combining strong intelligence with much higher output speed than most competitors.
The lesson: The best model is not always the smartest model on a benchmark. For many real-world use cases, the winning model is the one that is smart enough, fast enough, and cheap enough to use at scale. Gemini 3.5 Flash looks like a very strong example of that tradeoff.
The caveat: Benchmarks compress a lot of complexity into one score. The Artificial Analysis Intelligence Index is useful, but it does not fully capture reliability, tool use, long-context performance, coding quality, reasoning depth, cost, latency, or how a model behaves in actual workflows.
AI Is Starting to Change What Counts as Mathematical Work
⚡ Bottom line Nature’s feature shows AI moving from calculator-like assistance toward research collaboration in mathematics, including surprising solutions to long-standing Erdős problems.
💡 Why it matters If general-purpose models can make useful mathematical connections without being built only for math, science work may change faster than institutions can update peer review, authorship, and verification norms.
🔎 What it means The near-term opportunity is not replacing mathematicians. It is building workflows where humans use AI to search, conjecture, test, formalize, and then rigorously check ideas that would otherwise stay hidden.
Nature’s research feature centers on a striking example: Liam Price, who has no formal mathematics training, used ChatGPT to help solve Erdős problem #1196 in a way specialists found genuinely surprising. This was not just a brute-force calculation. The system helped surface a connection between number theory and probability that mathematicians said cut across familiar habits of thought.
That is why the story feels bigger than one problem. Researchers quoted by Nature describe a fast shift from AI as a tool that summarizes or automates toward AI as a partner that can suggest proof strategies, find unexpected routes, and expose gaps in human mathematical taste. The models are still limited, especially on long proofs, and much of the output still needs careful human checking.
The tension is exactly the point. If AI can generate plausible mathematics faster than journals, referees, and research groups can verify it, the bottleneck moves from discovery to trust. The best labs will not be the ones that simply ask models for answers. They will be the ones that build rigorous human-AI loops for conjecture, proof, review, and replication.
The IT strategy every team needs for 2026
2026 will redefine IT as a strategic driver of global growth. Automation, AI-driven support, unified platforms, and zero-trust security are becoming standard, especially for distributed teams. This toolkit helps IT and HR leaders assess readiness, define goals, and build a scalable, audit-ready IT strategy for the year ahead. Learn what’s changing and how to prepare.
