Since the last Digital Dips, things escalated. A guy built an mRNA vaccine with ChatGPT 5.4 to save his dog from cancer. Not in a lab. At his kitchen table… Meanwhile, Sam Altman once said AI would let one person build a billion-dollar business. That sounded bold. Now it sounds outdated. Because it already happened. Moving on, leaked documents from Anthropic about their upcoming model erased $20 billion in market cap. Just like that. A PDF can move markets now. And OpenAI? They shut down Sora ahead of their next release. Not because they’re scared. Because they’re serious. Companies in trouble cut costs in silence. Companies that believe in what’s next burn a billion-dollar partnership with The Walt Disney Company in public to free up compute. Altman says the new model will “really accelerate the economy.” It’s the first one built on a completely new architecture since GPT-5. Exciting times are ahead of us. And they’re not waiting around.

On today’s menu: Meta’s push toward personal superintelligence, Google giving Gemma the keys to the open world, quantum computers circling crypto like a ticking clock, Zapier turning 9,000 apps into one connected nervous system, and OpenAI and Anthropic laying the groundwork for a new class of AI. Systems that go far beyond next-token prediction and start reasoning, acting, and improving on their own.

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.

Meta just introduced Muse Spark, launching a new era in AI with its multimodal reasoning model designed by Meta Superintelligence Labs. Known for its tool-use and visual chain of thought, Muse Spark introduces a "Contemplating mode" (multi-agent orchestration) and seeks to pave the way to personal superintelligence. While it excels in multimodal perception and health reasoning, it shows gaps in coding workflows. Muse Spark is the foundation for future growth, as Meta outlines both current capabilities and future plans.

Read it yourself

Zuckerberg's risk paid off. By betting $14.3 billion on a 28-year-old who had never trained a frontier model, Meta shipped Muse Spark in just nine months. The brainchild of a CEO who specialized in data labeling shows impressive successes in benchmarks dependent on data quality. Yet it falls short in areas like coding where architecture innovation matters. This raises an intriguing question: Will future models close this gap, transforming Meta from a solid contender to the cream of the crop? If Meta cracks coding, that hefty investment might just redefine their position in the AI race.

Google's new research reveals that future quantum computers could crack Bitcoin's encryption using far fewer resources than previously thought. This discovery puts the cryptographic security of cryptocurrencies on a collision course with quantum advancements, urging a shift to post-quantum cryptography by 2029. Google's study warns of potential breaches within nine minutes with a 41% success rate, indicating cryptocurrencies need urgent security overhauls.

Read it yourself

Google's white paper is a wake-up call. It signals that quantum threats to cryptocurrency systems are not a distant theory but an impending reality. By dramatically cutting the quantum resources required, Google has effectively shortened the timeline for quantum computers to become a real threat. Cryptocurrencies now face a stark choice: adapt their security protocols or risk becoming obsolete. If there's any silver lining, it's that the race to secure digital currencies has now hit the fast lane. Solutions need to accelerate if they aim to outrun quantum capabilities.

Google just launched Gemma 4 under the Apache 2.0 license, giving developers complete freedom to modify and distribute without restriction. This shift comes after previous complaints about the usability of earlier versions under Google's custom license. Now, even the 26B model can run on a single consumer GPU, making advanced AI capabilities more accessible than ever. This might be the cleanest open offer in the competitive open model landscape, setting Google apart from rivals like Meta's Llama.

Read it yourself

Google's licensing switch for Gemma 4 is a strategic masterstroke. The shift from a restrictive license to Apache 2.0 opens the floodgates for startups and independent devs to use advanced AI without legal headaches. Where competitors still set boundaries, Google just handed AI's keys to everyone. It's no longer just about technical prowess; it's about who empowers developers most effectively. And with hardware efficiency offering accessible high performance, Gemma's appeal is unmatched. The game now is to make Google's AI your go-to starting point, because why wouldn't you?

Claude Managed Agents is now in public beta, offering APIs for rapid deployment of cloud-hosted agents. This suite lets users skip months of infrastructure work, moving from prototype to production in days. Early adopters have reported deployment speeds up to ten times faster, signifying a significant reduction in operational overhead. The platform stands to disrupt existing agent orchestration startups, offering built-in orchestration, security, and analytics tools.

Read it yourself

This launch sent shockwaves through the agent orchestration market. Shipping agents used to mean months of painstaking infrastructure work. Now, Anthropic has made it practically effortless. This shift could obsolete many startups thriving on agent reliability. The real narrative here is how Anthropic's offering mirrors Amazon’s game plan with their cloud services, delivering what businesses truly crave: infrastructure that just works, letting teams focus on outcomes, not overhead.

Two prominent AI labs, OpenAI and Anthropic, are on the cusp of breakthroughs with their latest models, GPT-5.5 (codenamed "Spud") and Claude Mythos. Rumors around GPT-5.5, suggest a leap in AI capability, with features like adaptive user intent understanding and extended long-term reasoning. At the same time, Anthropic has unveiled Claude Mythos, an AI model that autonomously finds security vulnerabilities. Both OpenAI and Anthropic are on the fast track to developing increasingly autonomous systems that challenge traditional computing limits. This shift isn't just about better chatbots—these models can automate complex tasks and even create new AI models. The implications for security, control, and work are immense.

What fascinates me most is not that both labs are racing. It is that they are slowly becoming independent from each other closing the loop. Building models that can help design new models. Systems that improve systems. That is no longer “next word prediction”. That is recursion. And recursion, at scale, is how you move towards AGI. Maybe even ASI. But here is the strange gap. On one side, users complain that ChatGPT once got a date wrong. On the other side, AI models are quietly finding software vulnerabilities that security experts used to make very good money fixing. It is like arguing about a calculator making a typo, while in the next room it is solving equations that used to take a research team a month. We are not having the same conversation. And notice this: this is not a product launch. It is a controlled deployment of systems considered too powerful to release freely. Anthropic is not releasing Mythos widely only to a small group of organizations. OpenAI has indicated that only a distilled version of Spud will be available, because the full system is simply too capable for society right now. That tells you everything. If the most advanced labs in the world are holding back their strongest models, it means the gap between what is possible and what is public is real. And likely growing. Add to that the fact that Chinese labs are reportedly only four to six months behind current frontier capabilities. By the end of this summer, you may have access to an extremely powerful AI model, perhaps operating at what feels like 80% of AGI, and free. Insane. The labs are closer to AGI than we might think. The questions of ethical control and economic disruption aren't far off, they're knocking. We need to think deeply about alignment before these systems fully integrate into our daily lives.

Zapier launched its SDK in open beta, offering a toolset that allows any coding agent to connect with over 9,000 applications seamlessly. The SDK eliminates the need for developers to manage OAuth setups, enabling quicker, integrated access to vast API capabilities. This streamlines workflows, making it easier to automate and scale across various platforms. As companies face increasing demand for integration, Zapier ensures that disparate systems talk to each other effortlessly.

Read it yourself

Zapier's SDK is transforming from a simple integration tool to the gatekeeper of company-wide connectivity. Instead of needing separate applications for different tasks, Zapier centralizes access to data and actions across platforms. This not only simplifies the technological landscape but democratizes access to complex integrations, allowing any developer or team to leverage comprehensive automation without the usual overhead. As more businesses recognize the value of such unfettered access, this approach sets a new standard for reducing complexity and amplifying productivity across the board.

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 fifty-four digital dip together. Forward this to someone who hasn’t realised AI is becoming the business.

Reading about agents is fun. Making them run your workflows is better. I help teams implement Claude Managed Agents the right way, connect 9,000+ apps through Zapier’s SDK, and turn AI news into clear action. If you want systems that actually work inside your organization, just reply to this email.

Looking forward to what tomorrow brings! ▽

-Wesley