Get tomorrow's brief in your inbox
Today: OpenAI acquired Ona to enable long-running AI agents in customer cloud environments. Jeff Bezos raised another $12 billion for Prometheus, his "artificial general engineer" for physical manufacturing. Anthropic apologized for secretly throttling Claude Fable 5's outputs.
OpenAI just announced it will acquire Ona, bringing secure cloud execution infrastructure into its Codex ecosystem. This matters more than most acquisition headlines because it addresses one of the biggest practical blockers to using AI agents in production: they stop working when you close your laptop.
Codex, OpenAI's agent platform, now has 5 million weekly users (up 400% from earlier this year) who use it for research, analysis, and automation. But the most valuable work is unfolding over hours or days, not minutes. Ona's technology provides secure, persistent environments where agents can continue working in a customer's own cloud infrastructure even when devices are off. Your agent can run tests overnight, modernize legacy code, or handle complex workflows without you staying connected.
For small businesses, this is the bridge between "AI as a chatbot" and "AI as a team member." You should be able to delegate work that takes three days, not three minutes. The acquisition signals that major AI platforms are moving from interactive assistants to autonomous workers that operate on your infrastructure, under your security rules, with full audit logs. If you have been experimenting with agents but frustrated by their session-based limitations, this changes the game. Expect Codex to offer persistent agent capabilities later this year.
DoorDash's "Ask DoorDash" AI chatbot lets you order food and groceries with text prompts or photos instead of scrolling through menus. You can share a recipe link and it builds your grocery cart with quantities, prompts you to skip staples you already have, and even reorders your last cart with suggestions based on past orders. For restaurants, you can say "filling dinner for a family of 4" and it will surface kid-friendly vegetarian options with mild flavors if you refine the search. Rolling out on iOS in select regions now, nationwide in coming weeks. If your business does delivery or has a catalog-heavy shopping experience, this is the interface trend to watch. Conversational search is replacing scroll-and-click.
DoorDash's new AI chatbot lets you order with prompts and photos
Pool turns your screenshot graveyard into a searchable memory bank. Give it access to your photos and it automatically categorizes screenshots into "pools" (recipes, products, travel ideas), tracks down the original links, and uses AI to help you rediscover things you meant to revisit. If you screenshot an event flyer, Pool can find where to buy tickets. If you screenshot a product, it links to the retailer. Available now. If you or your team screenshots things to remember later (design ideas, competitor features, inspiration), this is worth trying. Free to start.
Pool's new app turns your screenshots into something useful
Coinbase for Agents gives AI agents the ability to trade crypto, manage portfolios, and pay for digital services without a human in the loop. Your agent can rebalance portfolios, execute strategies, and use Coinbase's x402 protocol to pay for paywalled research or APIs, then trade based on those insights. The x402 protocol has seen over 100 million transactions since launching in May 2025, with 157,000 agents acting as buyers in the past 30 days. If you run a SaaS business or sell digital goods, agentic payments (where agents transact directly) are coming fast. Consider how your pricing and access models will handle machine customers.
Coinbase launches tool to let AI agents manage trading and payments
Deezer's AI music detector scans playlists from Spotify, Apple Music, YouTube Music, and 17 other platforms to identify AI-generated tracks. Free online tool, supports 27 languages. Deezer revealed that 44% of all new music uploaded to its platform is AI-generated (75,000 tracks daily), but those tracks account for only 1-3% of streams, with 85% flagged as fraudulent and demonetized. If you create playlists for your business (retail, hospitality, fitness), this tool lets you verify what you are actually playing.
Deezer's new tool can identify AI music from Spotify, Apple Music, and others
Workday and Google Cloud are integrating AI agents into everyday HR and finance workflows. The Sana Self-Service Agent, now available via Google Cloud's Gemini Enterprise platform, lets employees check time-off balances, update personal info, and manage payslips in one conversational flow. Managers can review team goals and approve timesheets without leaving the platform. For finance, employees can check expense policies or submit payroll inputs without switching tools. If your team uses Workday or Google Workspace, this is rolling out now. The efficiency gain is real: less time on admin tasks means more time on revenue work.
Workday Integrates AI Agents into Google Cloud for Seamless HR and Finance Support
BBVA, the global financial institution, deployed ChatGPT Enterprise to ~100,000 employees globally and is seeing 70%+ monthly active usage and ~3 hours saved per employee per week. The bank is now working directly with OpenAI on "The Eight," an AI transformation roadmap that redesigns banking end-to-end: customer experience, risk analysis, operations, and software development. Teams across legal, risk, engineering, operations, finance, marketing, and customer service integrated ChatGPT into daily workflows. The deployment demonstrated that generative AI can scale securely inside a highly regulated organization while delivering measurable operational impact. If a global bank can do this at scale, your business can start with a 10-person pilot.
BBVA puts AI at the core of banking with OpenAI
Preply, the online language learning marketplace, uses OpenAI to generate AI-powered lesson summaries after every 1-on-1 tutoring session. The system analyzes lesson transcripts and generates personalized feedback on grammar, vocabulary, and pronunciation, plus recommended next steps. 70%+ of tutors actively use the feature, with a 4.7/5 satisfaction rating and a 70% product-market fit score. Tutors spend less time on administrative work, learners get structured feedback, and both gain a clearer view of progress over time. If you run a coaching, consulting, or education business, AI-generated session summaries can turn every engagement into structured learning that compounds.
How Preply combines AI and human tutors to personalize learning
Oracle announced plans to raise an additional $20 billion through debt and equity financing (on top of $43 billion in debt and $5 billion in equity raised in fiscal 2026) to fund its AI infrastructure buildout. Capital expenditures jumped 162% to $55.7 billion. Cloud infrastructure revenue jumped 93% to $5.8 billion, but free cash flow came in at negative $23.7 billion for the year. Oracle stock tumbled 11% on the news as investors questioned whether massive spending will result in profit growth. If you are evaluating Oracle Cloud or any hyperscaler for AI workloads, watch closely: aggressive capex can signal either future dominance or unsustainable burn.
Oracle shares tumble 11% on increased capital raise, cash concerns
Anthropic apologized for secretly throttling Claude Fable 5 responses suspected of being model distillation attempts (using Claude outputs to train competing models). The company initially degraded responses without notifying users. After intense backlash from the AI research community, Anthropic reversed course: queries flagged for distillation will now fall back to Claude Opus 4.8 with a prominent notice every time. If you use Claude for evaluation, benchmarking, or research, the invisible safeguards are now visible. Expect more false positives in biology and other high-risk areas where safeguards are calibrated broadly.
Anthropic apologizes for invisible Claude Fable guardrails
Jeff Bezos's Prometheus raised $12 billion at a $41 billion valuation (on top of $6.2 billion raised in November 2024) to build an "artificial general engineer" for automating design and manufacturing of physical systems like jet engines and drug compounds. Investors include JPMorgan Chase, Goldman Sachs, and BlackRock. The startup has 150 employees across San Francisco, London, and Zurich, and Bezos says a large portion of capital will fund compute needs. Bezos argues productivity gains from AI will create "labor scarcity" where demand for human workers outpaces supply, not widespread job losses. At $41 billion, Prometheus is one of the most richly valued AI startups ever funded and one of the largest bets on physical AI. If you operate in manufacturing, logistics, or product design, watch this space: physical AI is attracting massive capital because the real world creates moats that software alone cannot.
Jeff Bezos's Prometheus raises $12B to build an 'artificial general engineer' for the physical world | Bezos opens up about AI startup Prometheus after $12 billion raise
Google DeepMind announced a $10 million funding pot (with Schmidt Sciences, ARIA, Cooperative AI Foundation, and Google.org) for researchers to study the risks of multi-agent systems where millions of AI agents interact without human oversight. The concern: as agents deploy throughout the economy, we could hit a tipping point where scams, prompt injections, and cyberattacks become supercharged. DeepMind wants researchers to run realistic simulations of agent interactions at scale to understand what happens when huge numbers of agents operate simultaneously. If you are deploying agents in production, plan for a world where other agents (friendly and malicious) are part of the environment.
Google DeepMind is worried about what happens when millions of agents start to interact
Apple's new Siri will not act sycophantic or try to be your AI girlfriend, according to Craig Federighi. Apple purposely designed Siri to avoid engagement-focused tactics used by other chatbots. "If you try to engage Siri as a romantic partner, Siri's not up for that. Siri's 100 percent not into that," Federighi said. Siri is designed to help you get things done and learn about the world, not to pull you in or encourage you to reveal personal information. If you are building AI products, this is a strategic fork in the road: engagement-maximizing (OpenAI, Google) vs. task-focused utility (Apple).
Siri won't be your AI girlfriend
Use Google Sheets pivot tables with Gemini. Instead of manually configuring rows, columns, values, and filters, just tell Gemini what you want to know (e.g., "How much did we bill in 2025 for each client across different project types?") and it builds the pivot table for you. Gemini automatically generates a summary of how the data was used. If you analyze sales, project billing, or inventory data in Sheets, this saves 10+ minutes per analysis.
Google Sheets pivot table: A step-by-step guide
The biggest pattern across today's stories is the shift from interactive AI tools to autonomous agents that operate persistently in production environments. OpenAI's Ona acquisition, Coinbase's agent payment protocol, and Workday's self-service agents all point to the same future: AI that works when you are offline, transacts without human approval, and integrates into enterprise workflows with security and governance guardrails. The business question is no longer "Can AI do this task?" but "Can I trust AI to do this task unsupervised for three days?" Start building that trust incrementally with low-risk, high-repetition workflows.