This is the fiftieth Digital Dips. 50 editions. Crazy! I started a little over two years ago. In that short time, we watched AI move from assistant… to operator. And I’m convinced that in the next two years, it will shift again. From operator to manager. On today’s menu: We now have AI that fakes emotion better than we do. Videos that cost 9 cents to produce. Websites slowly turning into APIs for AI agents. And even bots that hire other bots. If there’s one takeaway from this week, it’s this: we’re building a parallel world. An internet for AI agents. With it’s own economy. Where are ‘we’ in that future?

So grab your favorite snack, settle in, and let's dip into what's cooking. No time to read? Listen to this episode of Digital Dips on Spotify and stay updated while you’re on the move. The link to the podcast is only available to subscribers. If you haven’t subscribe already, I recommend to do so.

Fresh tech nuggets. Short, sharp, snackable.

Seedance 2.0 is now available for beta users on CapCut, marking a significant leap in AI video technology. This model raises the standard for motion, realism, and coherence, transforming what used to be a complex production into something you can achieve with a single prompt. In minutes, it delivers stunning, high-quality video outputs that could easily stand alongside premium content, blurring the lines between amateur and professional filmmaking.

Read it yourself

Seedance 2.0 impresses me most in how it handles physics in AI-generated video. I recently watched an AI-rendered football match that looked so real I had to read the comments to realize it was entirely AI. This leap in AI video doesn’t just democratize content creation. Another example I saw online came from the Formula 1 film starring Brad Pitt. The original shot from the film was expensive to produce, but with Seeddance it was recreated for just 9 cents. And instead of an F1 car, the clip showed an SUV with a cat behind the wheel… As more people get their hands on Seedance 2.0, we could see an explosion of high-quality user-generated content that people will genuinely believe is real. Remember that early clip of Will Smith eating spaghetti? Consider that ‘benchmark’ closed.

ElevenLabs rolls out Expressive Mode, bringing unprecedented emotional nuance to AI conversations. By integrating a conversational model with better turn-taking and advanced text-to-speech, voice agents can de-escalate and guide conversations with empathy in real time. Available in over 70 languages, this evolution narrows the gap between human and AI interactions, enhancing global customer experience.

Read it yourself

Emotionally expressive voice agents across 70+ languages represent a significant leap for customer-facing AI. Emotion has always been the missing piece between "technically correct" and "actually helpful." Expressive Mode from ElevenLabs is closing that gap. If AI can truly grasp tone and emotion in real time, we're stepping into a new era of customer experience. The distinction between AI assistants and humans just got even blurrier, potentially making AI better at handling frustrations than human counterparts. We’re officially in the era where AI cares more about your billing issue. Or at least fakes it better than the human customer rep do.

MiniMax unveiled M2.5, an open-source AI model geared for real-world productivity. It outperforms heavyweights like Claude Opus and GPT-5.2 on key benchmarks, particularly in tool use and multi-codebase tasks. At just $1 per hour for 100 tasks per second, it's a game-changer for scaling AI efficiently within the enterprise. This makes deploying cost-effective, smarter agents a reality, transforming AI infrastructure into a practical tool for production work.

Read it yourself

Normally, it was assumed that Chinese open-source models lagged about 4 to 6 months behind American closed-source models. Minimax 2.5? At most two months. That’s unprecedented, and a real shift from how things used to be. No doubt this release raised a few eyebrows in San Francisco. The real story here is MiniMax's ability to compete with giants on critical benchmarks while keeping costs down. This is about making AI scalable and practical for everyday use. The open-source approach combined with real-world productivity focus changes the AI landscape significantly. M2.5 offers a viable path to building expansive AI applications without breaking the bank. No organization has any excuse not to start integrating AI. There simply aren’t any left. With this model, you can continue developing everything in-house, without dependencies on China or the US, and still transform your company.

In just three months, the Gemini 3 Deep Think upgrade nearly doubled its performance on the challenging ARC-AGI-2 benchmark, scoring 84.6%. This leap was accompanied by an 82% reduction in cost per task, now down to $13.62. Google's fine-tuning of inference-time optimization is reshaping the landscape of AI reasoning, setting a pace others are struggling to match.

Read it yourself

Let's talk about the real story here: efficiency trumps brute force. Google's focus on inference optimization is a game-changer that shifts the reasoning race. No longer is AI progress about throwing more compute power at the problem; it's about smarter, leaner algorithms. With Deep Think now mainstreaming capabilities once thought exclusive, the gap between AI leaders and followers is widening rapidly. Expect the benchmarks to shift again, Google just redefined the playing field.

WebMCP is in early preview, offering a standardized way for AI agents to interact with websites without seeing the UI. Sites register structured tools directly in the browser, allowing agents to perform tasks quickly and accurately. This development is set to revolutionize how AI agents navigate websites, improving reliability and efficiency by replacing traditional methods with structured APIs. As more sites adopt this, we might see a shift in web design priorities.

Read it yourself

Google and Microsoft have turned every website into a potential API for AI agents with WebMCP, creating second-order effects that are hard to ignore. Where agents previously stumbled through guessing interfaces, they now have a clear roadmap for task completion. This has massive implications for site competition and optimization. In the near future, those who don't optimize for AI agents the way they once did for search engines may find themselves at a disadvantage. The landscape of the web is about to change, and it's ushered in by bots that already dominate 51% of web traffic.

A cheesy selection of three tools and one tasty rabbit hole.

A roundup of updates that are too cheesy to ignore.

This was it. Our fiftieth digital dip together. Forward this to someone who thinks AI is just a smarter search bar.

Maybe you’re reading this and thinking: my website needs to work for AI agents. Not just for humans. But where do you even start? What does that mean for your content, your tools, your data, your team? If you feel the shift but don’t yet have the plan: reply to this email. I’ll help you think it through.

Looking forward to what tomorrow brings! ▽

-Wesley