Founders, funding, and the build cycle — narrated.
Daily startup briefing — fundraising, founder stories, product launches, and the venture landscape — summarized and read aloud.
Stella Startups Brief — LawX takes €7.5M for European legal OS, Greenpixie cuts AI energy waste, and a Danish enzyme play on alcohol intolerance
Storyflo Daily·3 minWelcome to Storyflo Daily Startups. The European LegalTech raise of the week: per TechEU, Berlin-based LawX closed €7.
Today's curated set
Storyflo's agent monitors thousands of partner publishers and promotes stories into the trending feed when they get multiple sends or operator curation.
Storyflo Daily·3 minStella Startup Brief — Mistral acquires Emmi AI, NEX Health raises €1M for hospital infection AI, and European foodtech rebuilds on fundamentals
Welcome to Storyflo Daily Startups.
Storyflo Daily·3 minStella Startups Brief — Viktor hits €12.9M ARR in 10 weeks, Corti beats OpenAI on medical STT, and AVIAN's industrial fire-prevention play
Welcome to Storyflo Daily Startups.
Anthropic launches Claude Opus 5, a cheaper AI model for coding, agents and enterprise workflows
Anthropic's new AI model, Claude Opus 5, is designed for enterprise workflows and coding tasks, offering nearly all the intelligence of its top-of-the-line model, Fable 5, at half the cost. Opus 5 is priced at $5 per million input tokens and $25 per million output tokens, making it a more economically viable option for businesses. The model is particularly strong in coding and knowledge-work evaluations, setting new state-of-the-art marks on benchmarks like Frontier-Bench and GDPval-AA. It surpasses Fable 5's best result on OSWorld 2.0, a computer-use benchmark, at just over a third of the cost. While Opus 5 remains behind rival models in certain domains, such as cybersecurity tasks and biology research, it excels in bounded tasks with specific outcomes, where its efficiency and cost-effectiveness shine. Anthropic emphasizes that Opus 5 is ideal for daily use, while Fable 5 is better suited for long-horizon, autonomous jobs. The company's charts highlight performance at a given cost rather than peak performance alone, and early customers have reported significant improvements in efficiency, accuracy, and speed. For example, Harvey, a legal AI company, achieved similar performance to Opus 4.8's maximum-reasoning mode while generating 26% fewer tokens on average. Anthropic's business model is centered around API and enterprise usage, where inference costs have become a board-level line item. The company is selling a behavioral story, showcasing Opus 5's ability to verify its work and iterate until it succeeds, which is a crucial capability for enterprises.
Google renames NotebookLM to Gemini Notebook and expands code execution to Pro users
Google’s AI research assistant just got a new name—Gemini Notebook. It’s the same tool people have been using as NotebookLM, but now it sits under the Gemini brand, which Google says better reflects how it’s already woven into things like the Gemini app and Search. The rename doesn’t change the core experience; you still get the same AI‑powered note‑taking, summarizing, and brainstorming features you’ve come to rely on. What’s new is that the code‑execution capability, which lets the notebook run snippets of Python and other languages, is now opened up to Gemini Pro subscribers. Previously that feature was limited to a smaller group of testers, so Pro users can finally experiment with live code right inside their notes without hopping over to a separate IDE. It’s a handy way to prototype ideas or check results on the fly while you’re drafting a report or planning a project. Google says the move is about making the assistant feel more like a natural extension of its broader ecosystem. By aligning the notebook with Gemini, they’re signaling that the same underlying model powers everything from search suggestions to the dedicated Gemini app, so the experience stays consistent across the board. For anyone already using the notebook, the rename is mostly cosmetic, but the expanded code support should make the tool more useful for developers and data‑curious folks alike.
OpenAI's Codex update lets agents build interactive enterprise workspaces via Sites and role-specific plugins
Agentic AI is moving rapidly from the developer terminal to the corporate world.
Microsoft launches MXC, an OS-level sandbox for AI agents, with OpenAI and Nvidia already on board
For the past two years, the technology industry has raced to make AI agents more capable — teaching them to write code, navigate software interfaces, manage files, and orchestrate multi-step workflows…
Microsoft debuts Surface RTX Spark Dev Box to run large AI models without cloud costs
Microsoft just rolled out a new Surface desktop they’re calling the RTX Spark Dev Box. It’s a compact machine built around Nvidia’s latest Blackwell‑based RTX Spark chip and a massive 128 GB of unified memory. In plain terms, that means you can load and run AI models that are over a hundred billion parameters right on your desk, without having to fire off API calls to the cloud. The hardware delivers about a petaflop of AI compute, which is enough to keep a large model busy for hours on end. The big win here is the memory architecture. By sharing a single pool of RAM between the CPU and GPU, the box can hold the huge key‑value caches that big models need for long contexts—something that would normally require a specialized cloud GPU with high‑bandwidth memory. Microsoft also tweaked Windows to make the most of that setup, adjusting how the OS allocates pages and schedules work across the different cores so the GPU doesn’t starve the rest of the system. Thermal design got a creative twist, too. The top of the chassis is 3‑D printed metal with angled vents that guide airflow efficiently, letting the system stay quiet even under sustained load. It runs at roughly a hundred watts, which is modest for a desktop but enough to keep a model training or serving all night without overheating. Out of the box the machine is ready for developers. Windows 11 Pro boots into a clean dark theme with developer mode on, PowerShell as the default shell, and the Windows Subsystem for Linux pre‑configured for GPU work. VS Code, Git, Python, Node, and Copilot are already installed, along with Microsoft’s AI tooling for model conversion and deployment. The idea is to cut the setup time to zero, so you can start coding the moment you open the box. Microsoft is positioning this as a way to cut down on unpredictable cloud bills. The plan isn’t to replace Azure entirely—once you’ve prototyped locally you’ll still push to the cloud when you need scale—but giving developers a predictable, fixed‑cost hardware option for the bulk of their work. It’s a bold move for a company that makes a lot of money from cloud services, but it could be a real relief for anyone tired of watching their GPU spend rack up on every inference call.
Could AI give us more time for creativity? Maybe if you’re an entrepreneur
At Toronto Tech Week a mixed crowd of tech workers, academics and founders dug into what AI might actually do to our daily grind. The big idea on the table was that if AI can take over the repetitive bits, we could end up redefining what “productivity” looks like and free up more space for creative work. But the panel made it clear that the extra time won’t be handed out evenly – it’s likely to favor those who already have the flexibility to shape their own schedules. A few of the tech leaders argued that automating admin tasks will hand us back hours for imagination, yet early research suggests the opposite: the faster you finish, the more you get asked to do. The conversation shifted to the emerging role of “managing agents” – basically overseeing AI tools that handle your inbox, calendar and other chores. That extra layer of oversight could actually add a new kind of drudgery, making self‑employment look increasingly appealing. For freelancers and entrepreneurs, the promise is that AI assistants could coordinate the mundane while they focus on the parts of the job they love. The panel also touched on the darker side of AI rollout. Companies have already used AI as a pretext for layoffs, blaming job cuts on technological disruption rather than over‑hiring during the pandemic. Some executives see AI as a convenient scapegoat, while others stress that the fear of AI stealing jobs is largely imagined. What’s needed, according to the moderators, is a broader cultural shift about what counts as work, plus safeguards to stop AI from just extending the workday. Labor unions and upcoming policy moves, like a new AI‑and‑Labour advisory council, could help keep the balance in check. Finally, the discussion highlighted that Canada is lagging behind the U.S., U.K. and Israel in adopting “agentic” AI, so any transformation of work will take longer than a software update. The consensus was that while AI will inevitably reshape how we relate to our jobs, the change will be gradual. For now, entrepreneurs seem best positioned to reap the creative benefits, while the rest of the workforce watches and waits for the larger shift to settle in.
Version One Ventures raises $108 million USD across two new funds
Version One Ventures, the Vancouver‑based early‑stage fund that’s been backing deep‑tech startups for a while, just closed two fresh funds totalling about $108 million USD. One is a $78 million fifth fund, and the other is a $30 million third Opportunities Fund. The firm says it’s shifting its lens toward AI infrastructure, robotics, physical AI, and biotech, and it’s also looking beyond North America to places like India and Africa. Boris Wertz, the founding partner, notes that Canada’s early‑stage scene hasn’t been as lively lately, which is why the firm’s Canadian portfolio has shrunk from roughly half of its investments to about a tenth. They’ve still got some big names in their rear‑view mirror—Ada, Dapper Labs, Jobber, Clio, Top Hat, Wattpad—and they famously turned a modest $2 million stake in Coinbase into a massive return. The broader picture lines up with a recent study that found almost half of Canada’s “high‑potential” startups are actually based in the United States. That trend, combined with the difficulty of attracting money to deep‑tech projects, explains why Version One is casting a wider net and chasing opportunities in emerging ecosystems. All in all, the fund’s new capital signals a clear pivot: double down on cutting‑edge tech and expand the geographic hunt, while still keeping an eye on the occasional Canadian gem that might pop up.
Barcelona’s Zazume raises €2.5 million to scale its AI-powered rental management platform
Zazume, a Barcelona-based PropTech startup that digitises the entire residential rental lifecycle, has closed an investment round worth €2.