Welcome, humans.

Ryan Singer had a very good complaint about Jira, Linear, and every other work tracker: they do not show the budget behind the work. Who authorized this task? What capacity is it spending? What did we choose not to do instead?

That feels like the next useful AI benchmark: can the model estimate the token budget required to complete a task, translate that into cost, and map it to rough real-world time? Basically, can it understand its own token burn rate per hour? Forget “can it pass the test.” Show me the AI that can say, “This will cost about $14.80 and three human-equivalent hours.” Now that is AGI.

Here's what happened in AI today:

😼 Google released Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber, splitting cheaper agent workflows from restricted cyber-defense use.
📰 OpenAI turned ChatGPT into a full business platform with ads, small-business agents, and a 3.2 GW Georgia data center.
📰 OpenAI and Anthropic lobbied Washington to restrict Chinese open-weight AI models.
📰 Nvidia's Jensen Huang pushed back, arguing those models aren't a threat worth blocking.

...and a whole lot more that you can read about here.

😺 Google Just Split Gemini Into Three Very Different Personalities

Not every AI model should be the same model. That's the idea behind Google's latest release: instead of one do-everything Gemini, the company just shipped three specialized versions built for three very different jobs.

Here's what happened:

Gemini 3.6 Flash is the new all-purpose workhorse, better at coding and document analysis while using 17% fewer output tokens (the units AI usage gets billed by) than its predecessor. It's also cheaper: $1.50 per million input tokens and $7.50 per million output tokens.
Gemini 3.5 Flash-Lite is built for speed: 350 output tokens per second, aimed at high-volume jobs like scanning documents or powering search.
Gemini 3.5 Flash Cyber is a locked-down model that only governments and vetted partners can access, built to find and patch cybersecurity holes before hackers do.

How to try it: 3.6 Flash and 3.5 Flash-Lite are live now in Google AI Studio, the Gemini Enterprise platform, and the regular Gemini app. Flash-Lite is also rolling into Google Search.

Why this matters: For most of AI's short life, "better" meant "bigger, smarter, one model that does everything." Google's move argues the opposite: agents (AI systems that complete multi-step tasks on their own) work best when the model is built for the job, not maximally powerful at every job. A cheap, fast model that's good enough beats an expensive genius model that's overkill for scanning receipts.

The cyber model is the more interesting bet. AI has gotten good enough to find security holes faster than humans can patch them, so Google built a specialized model to fight that fire, then locked it behind a government-and-partners-only pilot instead of releasing it to everyone. That's a tell: even Google isn't fully comfortable putting its best vulnerability-hunting AI in just anyone's hands.

Our take: This is Google quietly admitting the "one model to rule them all" era is ending, and the real fight is now over who builds the smartest lineup. Open question worth sitting with: if AI can find security holes faster than we can fix them, does gatekeeping one cyber model matter once a rival ships a public version?

FROM OUR PARTNERS

Success in AI is no longer about generating cool, one-off videos. It is about building dynamic, context-aware media systems at scale. To help you build the next big thing, the Google for Startups team just released a new Technical Guide to Generative Media.

Inside, you'll discover how to integrate DeepMind models like Gemini Image, Gemini Omni, Veo, and Lyria into production workflows. You'll also get a technique library and expert advice on the orchestration tools that help make outputs repeatable.

Get the guide and turn your creative vision into reality.

🎓 AI Skill of the Day: Build a Claude Design System Before You Build Slides

Claude Design gets generic when you start with a blank template and hope one heroic prompt fixes everything. Jeff Su’s 14-minute tutorial shows a better approach: prepare the rules once, then reuse them across every deck.

His workflow uses three inputs:

A DESIGN.md file that defines colors, fonts, spacing, buttons, and cards.
A Claude Design system that turns those rules into reusable components and sample slides.
A presentation template that controls layouts such as title slides, section dividers, and two-column pages.

Start with a public DESIGN.md example, then ask ChatGPT to remove proprietary brand details while preserving the useful structure. Upload the cleaned file when creating your Claude Design system. Once Claude produces a template, give specific feedback and ask it to save those standing instructions in a CLAUDE.md file so future decks inherit every correction.

Our favorite part: you stop reprompting the same brand rules and start improving one design system that compounds over time.

Clean the attached DESIGN.md file for use as my own design system.

Remove or replace all proprietary brand names, logos, slogans, product references, and distinctive copyrighted language. Preserve the file’s useful structure, including its rules for colors, typography, spacing, layouts, buttons, cards, and visual hierarchy.

Rename the result: [YOUR DESIGN SYSTEM NAME]

Return the complete cleaned file in Markdown. Do not summarize it.

Want more tips like this? Check out our AI Skill of the Day Digest for July.

🍪 Treats to Try

Want your tool in front of 700K+ AI-curious readers? Advertise to 700K+ readers here!

Language Builder teaches you how to, quite literally, build your own language model! (from Anthropic Cowork creator Felix Rieseberg).
Synthesia Roleplay Sessions lets teams practice sales calls, performance reviews, and customer complaints with AI avatars that act out realistic workplace scenarios —pricing by demo.
Samsung's new Galaxy foldables ship with Gemini Notebook pre-installed, making Google's AI note and research tools part of the default premium-phone workflow —device purchase required.
Meta StoryKit generates customized children's stories with characters, settings, lessons, and music —testing only, pricing not public.
Poolside Laguna S 2.1 gives developers an open-weight coding model for long-horizon agentic software work —from $0.10/$0.20 per million input/output tokens.
Block Buzz gives teams a free, open-source workspace where humans and AI agents can share channels, repositories, workflows, and portable identities —free/open-source.
Substack's Pangram scanner lets readers estimate whether posts, notes, replies, and comments over 100 words were AI-assisted —rolling out inside Substack.
Gemini Task Automation can trigger recurring and context-aware actions across 40+ apps as Samsung brings the feature to its new foldables —device and Gemini availability apply.
Anthropic Economic Index lets you ask Claude which jobs use AI most and what tasks people are automating; open the Connectors menu, find Anthropic Economic Index in the directory, enable it, then ask your question in any chat —free to try.

New from The Neuron: AI Explained… When Will Robots Be Better Surgeons Than Humans?

In our newest podcast episode, Inner Logic cofounder Mathias Unberath explains how surgical robots can learn anatomy, recover from mistakes, and gradually automate precision procedures without removing human surgeons from the operating room.

Watch and/or Listen on: Spotify | Apple Podcasts | YouTube

📰 Around the Horn

OpenAI turned ChatGPT into a full business platform, adding ads, a small business agent program, and a 3.2 GW data center buildout in Georgia.
OpenAI and Anthropic aligned in Washington against powerful Chinese open-weight models, while Nvidia's Jensen Huang defended them as demand drivers.
Frontier AI labs pushed Washington to restrict Chinese open-weight models, while open-source builders and Nvidia's Jensen Huang argued the models aren't inherently dangerous.
Nearly 200 utilities and data-center companies signed Trump's voluntary pledge to shield ratepayers from AI power costs, though enforcement remains weak.
Reddit and major publishers are reconsidering Google content access as AI answers cut referral traffic and weaken the old search-for-clicks bargain.

FROM OUR PARTNERS

AI innovation is accelerating, but not every breakthrough belongs on your roadmap. Join NiCE experts Neeraj Verma and Andy Traba to learn which CX AI innovations matter most, how they work together, and how orchestrating intelligence can drive measurable business outcomes across the enterprise.