Demis Hassabis on curing diseases, DuckDuckGo's 30% surge, and OpenRouter hits $1.3B

Something quietly shifted this week - and it might be the clearest signal yet that ordinary users are starting to push back on AI being forced into every corner of their digital lives. DuckDuckGo's install numbers don't lie. Meanwhile, the person arguably closest to building AGI just gave one of his most candid interviews in years. Let's get into it.

How many times did OpenRouter's usage grow in just six months before its latest funding round?

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

Demis Hassabis, CEO of Google DeepMind, sat down for a rare wide-ranging interview this week - and it's worth your full attention. The man who built AlphaFold and helped crack protein structure prediction at scale is now talking openly about what comes after that.

Hassabis frames the next phase of DeepMind's mission as using AI to compress decades of biological and medical research into years. His argument is specific: AI can model biological systems at a fidelity we've never had before, and that opens up drug discovery pipelines that would have taken entire careers to navigate. He also addresses AGI directly - cautious, but notably less dismissive than he's been in past interviews.

For anyone tracking the AI healthcare space, this interview is essential reading. The gap between "AI assists doctors" and "AI discovers cures" is closing faster than most people realize.

Read the full story →

OpenRouter just closed a $113 million Series B led by CapitalG, Google's independent growth equity fund, pushing its valuation past $1.3 billion. A year ago it was valued at roughly half that. The rocket fuel? A 5x growth in usage over just six months.

OpenRouter sits between developers and dozens of AI models - OpenAI, Anthropic, Mistral, and more - through a single unified API. Developers stop rewriting integrations every time a new model drops. Instead, they swap models with a config change. For startups iterating fast, that's a genuine superpower.

This is what the multi-model future actually looks like in practice. Not one dominant model, but a fluid layer where you pick the best tool for each task. The fact that usage grew 5x in six months tells you how many developers have already internalized this approach.

Speaking of building fast - if you're spinning up a new project and need a landing page or site live in under a minute, 60sec.site uses AI to get you there without the usual setup headaches. Worth a bookmark.

Read the full story →

Google made a bold call at I/O 2026: replace traditional blue-link search results with AI agents by default. The backlash was swift. DuckDuckGo app installs jumped 30% almost immediately, as users actively sought an alternative that would just show them links.

There's a meaningful difference between users who choose AI-generated answers and users who have them forced on them. Google seems to have underestimated how many people in the second camp exist. DuckDuckGo's pitch - private search, no tracking, no AI summaries unless you want them - is suddenly more appealing than it's been in years.

This connects to a broader pattern worth watching. Remember when we covered Google's AI Search overhaul and the 'zero click' concern for the web? The user revolt is now showing up in the install data.

If you're curious about how AI regulation and consumer pushback could reshape search long-term, this is a story to keep an eye on.

Read the full story →

ClickUp, the nine-year-old productivity startup, just made one of the most blunt AI-for-headcount swaps we've seen from a software company: it laid off hundreds of employees and is replacing them with thousands of AI agents. Not automating tasks around staff. Replacing staff with agents.

Hundreds of people out, thousands of agents in. That ratio is the whole story. ClickUp is betting it can operate at higher output with a dramatically leaner human team by leaning into AI agents for everything from customer support to internal operations.

Whether this works as a business strategy is genuinely an open question. But as a signal for the future of work, ClickUp is planting a flag that others in the productivity and SaaS space will be watching closely. The pressure on other CEOs to follow suit - or justify why they haven't - just went up.

Read the full story →

Human Archive, a startup founded by UC Berkeley and Stanford researchers, has a different take on the robot training data problem. Instead of synthetic data or expensive lab setups, they're paying gig workers in India to wear camera-equipped caps and sensor devices while going about everyday physical tasks - walking, carrying, picking things up.

The core insight is simple but underappreciated: robotics models need real-world physical data at massive scale to generalize properly. Simulated environments only get you so far. Human Archive is essentially turning India's large, flexible gig economy workforce into a real-world motion capture operation for the next generation of humanoid robots.

It raises some interesting ethical questions about data labor and compensation - but as a technical solution to one of physical AI's hardest problems, it's genuinely creative. Robotics labs are racing to acquire exactly this kind of embodied data, and Human Archive is betting it found the right supply chain.

Read the full story →

The answer is 5x! OpenRouter's usage grew five times in just six months leading up to its Series B raise - which is precisely what convinced CapitalG to lead the $113M round and push the valuation past $1.3 billion. That kind of growth rate in a six-month window is rare even by AI startup standards.

Are you using a tool like OpenRouter to switch between AI models, or do you mostly stick to one? I'm curious whether the multi-model approach is actually part of your workflow yet - hit reply and tell me. I read every response!

That's it for today - see you tomorrow with more. And if you want to catch up on anything you missed, the full archive is always there for you at Daily Inference.