Get tomorrow's brief in your inbox
Today: Anthropic's Claude hit 1 million daily signups after refusing Pentagon surveillance terms, pulling ahead of ChatGPT in downloads. Zapier's week-long hackathon jumped company AI adoption from 10% to 50%, proving hands-on building beats mandates. Salesforce shipped Agentforce for telecom, reclaiming 300+ hours per week for enterprise teams.
Anthropic's consumer growth surge shows what happens when you pick principles over Pentagon contracts. After CEO Dario Amodei refused to let the government use Claude for mass surveillance of Americans or fully autonomous weapons, the Defense Department marked Anthropic as a "supply chain risk" (a label typically reserved for foreign adversaries). The result? Claude became the #1 app on the U.S. App Store, now seeing 1 million signups per day and 11.3 million daily active users (up 183% from January). U.S. downloads continue to beat ChatGPT (149,000 vs 124,000 daily), and web traffic jumped 43% month-over-month.
For small businesses, this matters beyond the headline drama. Microsoft, Google, and Amazon confirmed Claude remains available to all non-defense customers through their cloud platforms (Microsoft 365, GitHub, Azure, Google Cloud, AWS). If you've built workflows around Claude or were considering it, you're clear to keep using it. The Pentagon designation only blocks DoD contracts, not commercial access.
The broader lesson: AI model competition is heating up, and differentiation increasingly comes down to trust and brand positioning. When capabilities commoditize (all top models perform similarly), customers gravitate toward providers whose values align with their own. If you're evaluating AI vendors, look beyond benchmarks to policies, transparency, and how they handle ethical questions.
Claude's consumer growth surge continues after Pentagon deal debacle | Microsoft, Google, Amazon say Anthropic Claude remains available to non-defense customers
Salesforce Agentforce for Communications - Salesforce launched five prebuilt AI agents tailored for telecom companies, but the productivity gains apply to any service business juggling complex customer interactions. The platform includes a Billing Resolution Agent (handles disputes autonomously), Guided Selling Agent (turns field visits into upsell opportunities), and 24/7 customer service agents. Lumen Technologies reports reclaiming 300+ hours per week. One NZ saw a 4x increase in customer engagement over traditional channels. If you run a service business with field teams or billing complexity, this pattern (AI handling routine tasks while humans focus on high-value work) is worth exploring. Pricing not disclosed, likely enterprise-tier.
Telecom Companies Turn to AI Agents to Boost Revenue and Customer Loyalty
Google Gemini app beta for task automation - Google announced a beta feature for the Gemini app (launching on Pixel 10 and Samsung Galaxy S26 series) that lets you offload multi-step tasks to your phone with a long-press of the power button. Currently limited to food delivery, grocery, and rideshare apps. The hands-free automation stops as soon as the task completes, and operates in a secure window with notifications to keep you informed. If you're constantly juggling DoorDash orders or Uber rides during work hours, this could claw back 10-20 minutes daily. Watch for the U.S. and Korea rollout; broader app support should follow if adoption is strong.
Google Unveils Beta Feature to Automate Daily Tasks with Gemini App
OpenAI Codex Security (research preview) - OpenAI launched an AI security agent that analyzes project context to detect, validate, and patch vulnerabilities. Think of it as a security engineer that reviews your codebase. Meanwhile, Anthropic's Claude Opus 4.6 found 22 vulnerabilities in Firefox over two weeks (14 high-severity), proving AI can audit complex codebases faster than human teams. If you run a dev shop or maintain custom software, AI-powered security reviews are shifting from experimental to practical. Start budgeting for tools like this.
Codex Security: now in research preview | Anthropic's Claude found 22 vulnerabilities in Firefox over two weeks
Grammarly "Expert Review" controversy - Grammarly's AI feature offers writing feedback "inspired by" subject matter experts (including journalists and deceased professors) without asking permission. The Verge found their own editors listed as AI personas giving writing advice they never agreed to provide. Sources frequently linked to spammy copies or unrelated pages. If you use Grammarly's AI features, treat "expert" suggestions skeptically and verify sources. For businesses building AI products: this is a case study in how not to handle attribution and consent.
Grammarly is using our identities without permission
Zapier's hackathon-driven AI adoption - When GPT-4 launched, Zapier CEO Wade Foster called a company-wide "code red" and stopped the entire company for a week to have everyone (engineers, marketers, salespeople, finance) build something with AI. Result: AI adoption jumped from 10% to 50% in one week. Foster says fear of AI fades when people actually use it and learn the limitations firsthand. If you're struggling with AI adoption at your company, mandates don't work. Hackathons do. Give your team access, clear a week, and have everyone ship something small. The hands-on experience matters more than the memo.
Zapier now requires all new hires to be AI fluent, rating candidates on a four-tier scale from "unacceptable" to "transformative." They update criteria every six months because AI moves fast. Foster's hiring insight: hyper-curiosity is the trait that matters most. If candidates just say "I asked ChatGPT to rewrite this memo," they're not exploring the full potential.
How a week-long hackathon transformed Zapier's AI culture
Automate yourself out of a job (it's a good thing) - Foster says the fear of automating your own work is backwards. "I know of no CEO or leader where a person came to me and said, 'I have automated my entire job,' whose reaction is not, 'holy crap, show me how.' And I have nine other projects I'm ready to turn you loose on." If you can systematically automate work, you become one of the most valuable people in the company. Run toward that skill, not away from it.
How a week-long hackathon transformed Zapier's AI culture
Zapier AI model flexibility - Zapier lets you swap AI models (Claude, GPT, Gemini) across workflows without rebuilding integrations, avoiding vendor lock-in. Your marketing team can use Claude for long-form content while Sales uses GPT for call transcripts. When pricing shifts or one model leaps ahead, you switch in seconds. If you're building AI workflows, choose platforms that let you change providers easily. Building directly on one vendor's platform (like OpenAI's Frontier) creates maintenance nightmares when APIs change or better models launch.
Prevent lock-in with AI model flexibility on Zapier
Claude remains available through Microsoft, Google, AWS - Despite the Pentagon supply chain risk designation, Microsoft (365, GitHub, Azure AI Foundry), Google Cloud (Vertex AI), and AWS all confirmed Claude stays available for non-defense customers. If you're already using Claude through these platforms, nothing changes. Google also invested an additional $1 billion in Anthropic (adding to its $2 billion stake), and Anthropic now has access to 1 million Google TPUs for model training. The enterprise access is secure.
Google joins Microsoft in telling users Anthropic is still available outside defense projects
City Detect raises $13M for AI-powered urban monitoring - City Detect mounts cameras on garbage trucks and street sweepers, uses computer vision to spot building code violations (graffiti, illegal dumping, structural damage), then helps cities fix issues. They can process thousands of buildings per week vs. 50 manually. Dallas and Miami are clients. For small businesses in property management or municipal services, this shows how computer vision is moving from experimental to operational. If you manage buildings or work with cities, AI-powered compliance monitoring is coming.
City Detect, which uses AI to help cities stay safe and clean, raises $13M Series A
Cluely caught inflating user numbers - The a16z-backed startup that helps people cheat on job interviews (AI whispers answers during video calls) got caught walking back a $7M growth claim. This is why AI ethics questions aren't just philosophical. Tools that help people game systems (interviews, credentials, compliance) create trust erosion that hurts everyone. If you're building AI products, the "what could go wrong" question matters as much as the "what's possible" question.
An a16z-Backed Startup That Helps People Cheat on Job Interviews Just Got Caught in a $7 Million Lie
Zapier AI automation pattern - Instead of building workflows around one AI provider, use Zapier's AI by Zapier tool with model selection in a dropdown. Start with one model, but design workflows that let you swap models later. This takes 30 seconds longer upfront and saves weeks when you need to switch providers.
Evaluate AI vendors on values, not just benchmarks - When top models perform similarly, differentiation comes from trust. Ask vendors: What are your data retention policies? How do you handle customer data in model training? What's your stance on controversial use cases? The answers reveal whether you can build a long-term relationship.
Three patterns emerged today: AI adoption works through hands-on building (not mandates), vendor lock-in is avoidable if you design for flexibility, and brand trust increasingly drives model choice as capabilities commoditize. Watch for more enterprises requiring "AI fluency" in hiring (Zapier won't be the last), and expect continued fallout from the Anthropic-Pentagon split as other vendors position themselves on ethics questions. The hackathon model for driving adoption will spread beyond tech companies.