Get tomorrow's brief in your inbox
Today: Anthropic launched Claude Design for quick visual prototypes and caught the White House's eye with its cybersecurity-focused Mythos Preview model. AI coding productivity claims hit a reality check as data shows developers rewriting up to 90% of accepted AI-generated code.
If you're using AI coding tools like Claude Code, Cursor, or GitHub Copilot and feeling more productive, the numbers suggest you should look closer. New data from developer productivity analytics firms reveals a sobering pattern: while AI tools increase code acceptance rates to 80-90%, engineers end up revising that "accepted" code far more often than before, driving real-world acceptance rates down to 10-30% of what was originally generated.
Waydev, which tracks analytics for 50+ companies employing 10,000+ engineers, reports that code churn (lines deleted versus added) has spiked. GitClear found AI users averaged 9.4x higher churn than non-AI counterparts. Faros AI's March 2026 report showed an 861% increase in code churn under high AI adoption. The pattern is consistent: more code is being written, but a disproportionate amount isn't sticking.
The problem is developers are measuring the wrong thing. Token budgets and initial acceptance rates are inputs, not outputs. If the code you accepted Monday gets rewritten Thursday, your productivity didn't increase. The lesson for small businesses: don't track how much AI code your team accepts. Track how much of it survives two weeks later. Use AI coding tools, but validate that they're actually saving time, not just creating an illusion of speed.
Anthropic's Claude Design makes visual prototypes from conversation. Describe what you want ("prototype a meditation app with calming colors and clean layout"), and Claude creates an initial version you can refine with direct edits or requests. You can export to PDF, PPTX, URL, or send to Canva for further editing. It can apply your team's design system by reading your codebase and design files. Available now for Claude Pro, Max, Team, and Enterprise subscribers. This is for founders and product managers who need to go from idea to visual quickly without opening Figma. If you spend hours trying to mock up slides or one-pagers in PowerPoint, this could cut that to minutes.
Google Gemini Notebooks consolidate your project data into personal knowledge bases that sync across Google products. Click "New notebook" in the Gemini app's side panel, move relevant chats in, upload documents and PDFs, and provide custom instructions. When you ask Gemini questions, it draws from that curated information to give tailored responses. Free to use. If you run multiple projects and waste time searching for data spread across emails, docs, and chat threads, this could save you hours per week. The risk: notebooks get cluttered if you don't maintain them, so set a weekly cleanup routine.
Zapier alternatives worth evaluating: If Celigo feels too complex for your automation needs, Zapier compared itself to MuleSoft, Microsoft Power Automate, Boomi, Workato, Informatica, and Fivetran. Zapier's pitch: 9,000+ app connections, AI orchestration via Zapier Copilot, and enterprise controls (SSO, SCIM, audit logs). Not FedRAMP or HIPAA compliant, so healthcare and government contractors need to look elsewhere. Free plan available, paid plans from $19.99/month. If your team closes tabs the moment they see JSON, Zapier's no-code-first approach beats Celigo.
iPhone automations let you trigger actions based on time of day, location, or opening an app. Nested inside the Shortcuts app (iOS 14+), they work like mini Zaps: trigger + action. Set up monthly Venmo payments, auto-send texts when you arrive somewhere, or log work hours automatically. Not robust enough to run your business, but useful for personal productivity tasks. If you forget recurring tasks that can't be automated elsewhere, this is worth 15 minutes to set up.
Dairy Queen is rolling out AI chatbots in drive-thrus across the US and Canada, built by Presto (same company working with Carl's Jr., Hardee's, Taco John's). The chatbot takes orders correctly about 90% of the time and is designed to speed up service and upsell customers. Presto's AI may be assisted by human workers in locations like the Philippines, per a 2023 Bloomberg report. If you run a QSR or retail operation with high-volume, repetitive customer interactions, AI voice is ready for production use. The business case: faster service during peak hours and consistent upselling without employee fatigue. The catch: 90% accuracy means 1 in 10 orders needs human intervention, so staff your drive-thrus accordingly.
China's Meituan AI agent (Xiaomei) shows where commerce automation is headed. Users say "Order my usual lunch, but deliver it 20 minutes later today," and the agent interprets intent, applies preferences, and completes the transaction with zero screen interaction. Harvard Business Review calls it "orchestrator plus execution agent." The point isn't convenience, it's delegation. If you run a service business with repeat customers, start thinking about how you'd enable AI agents to act on your customers' behalf. The ones who make it easy for agents to book appointments, reorder supplies, or schedule service will win when this hits the US market.
AI training for employees: Companies chasing tools are falling behind. The smart ones build skills. Inc.com reports that focusing on foundational AI literacy (how to prompt, when to use AI, how to validate output) beats tool-specific training. Tools change every quarter. Skills compound. If you're training your team on "how to use ChatGPT," pivot to "how to evaluate AI output quality" and "when AI is the wrong tool for the job."
Cursor is raising $2B+ at a $50B valuation as enterprise adoption surges. The AI coding startup forecasts $6B annualized revenue by end of 2026, up from $2B in February. TechCrunch reports Cursor recently achieved slight gross margin profitability by introducing a proprietary Composer model and routing to cheaper models like China's Kimi. Individual developer accounts still lose money, but enterprise accounts are profitable. If you're an individual dev paying for Cursor, you're subsidizing enterprise customers. If you run a dev team and haven't evaluated Cursor versus Claude Code or GitHub Copilot, put it on your list.
OpenAI is shedding "side quests" and losing key talent. Kevin Weil (VP of AI for Science) and Bill Peebles (Sora team leader) both left Friday as OpenAI consolidates around enterprise AI. Sora was burning $1M/day in compute costs and shut down last month. OpenAI for Science (the Prism research platform) is being absorbed into other teams. If you were building workflows around Sora, it's gone. The pattern is clear: OpenAI is focusing on enterprise coding and its forthcoming "superapp." Consumer-facing experiments are being cut.
Anthropic is warming up to the White House after a two-month freeze. CEO Dario Amodei met with White House Chief of Staff Susie Wiles on Friday to discuss Claude Mythos Preview, a cybersecurity-focused model that can find vulnerabilities in browsers and operating systems. Anthropic briefed senior US officials and hired Trump-linked lobbying firm Ballard Partners. Apple, Nvidia, and JPMorgan Chase have already signed on to use Mythos Preview. The model is powerful enough to spark emergency meetings between US bank leaders and Fed Chair Jerome Powell. If you run a security-sensitive business, this is the model to watch. Private access only for now.
Sam Altman's World ID is expanding beyond orb-scanned crypto verification into Tinder, Zoom, Docusign, and concert ticketing. Tinder users who verify via World's iris-scanning orb get a "verified human badge" and five free boosts. Concert Kit reserves tickets for World ID-verified humans to fight scalper bots, compatible with Ticketmaster and Eventbrite. 30 Seconds to Mars and Bruno Mars are using it for upcoming tours. The pitch: prove you're human while staying anonymous via "zero-knowledge proof-based authentication." If you run events or services plagued by bots, this is worth evaluating. If you're creeped out by iris scanning, stick to government ID verification.
Use Google AI Mode's Canvas tool for custom trip planning. Describe your ideal trip, and AI Mode crafts an itinerary in the side panel with flight/hotel options and attractions on a map. Ask follow-ups to refine. It auto-saves progress in AI Mode history. Available now for everyone in the US. If you waste hours on travel research, this condenses it to one conversation.
Track individual hotel prices on Google Search (desktop and mobile). Tap the price tracking toggle when searching for a specific hotel, and you'll get an email alert if rates drop significantly during your dates. Available globally for signed-in users in English and Spanish. If you're planning travel in the next 3-6 months, turn this on for every hotel you're considering.
The gap between AI hype and AI reality is widening. Cursor is valued at $50B while data shows AI-generated code gets rewritten at 9x the rate of human code. OpenAI is cutting consumer experiments to focus on enterprise. Anthropic is pivoting to cybersecurity to win back government contracts. The pattern: AI companies are optimizing for enterprise buyers who pay for reliability, not consumer users who pay for novelty. If you're betting on AI tools for your business, bet on the boring ones solving real workflow problems (coding, design prototyping, security scanning), not the flashy ones making demos (video generation, consumer chatbots). The useful AI tools are the ones that save you time two weeks after you start using them, not two minutes.