Axis Robotics raised a $12M seed round led by Hack VC to tackle what could be the biggest bottleneck in Physical AI: data. The thesis is pretty simple. Robots can’t browse the internet to learn how to move. LLMs had the entire web to learn from. Robots don’t have that same foundation. @axisrobotics is building the data engine to help fill that gap. What they’re building: - Turn robot training into a browser-based game where anyone can teleoperate a virtual arm, and every completed task becomes usable training data, recorded on Base. - A mobile app for capturing real human movement and generating first-person data. - A pipeline that processes all of that information into model-ready datasets. - Then package and sell those datasets to robot manufacturers, foundation model teams and industrial automation companies, with names like Booster Robotics, Geely and Lotus already involved. The model is essentially B2B2C: The community contributes the data and earns from it. Enterprises pay for access to the resulting datasets. And the traction is already there: - 130K+ contributors - 4M+ trajectories The bigger opportunity is the data layer underneath Physical AI. If Axis can become the infrastructure that continuously supplies robots with the data they need to learn, the upside could be significant. That’s the bet they’re making. And honestly, it’s an interesting one to get in early on.

AP Collective is heading to Seoul for Korea Blockchain Week. If you'll be around, come say hi.

Bitcoin has billions of dollars of value sitting onchain, but most of that value isn’t doing much beyond being held. What if Bitcoin could become more useful without changing what makes Bitcoin valuable? That’s the direction @Starknet is exploring. Starknet is building infrastructure that allows Bitcoin to interact with a programmable environment, opening the door for BTC to be used across applications, DeFi and other onchain activities. And this goes beyond simply creating another representation of BTC. With strkBTC, Starknet is exploring how Bitcoin can be brought into this ecosystem while also giving users an option to use privacy features through its STRK20 infrastructure. You can think of it as adding another layer of utility around Bitcoin: BTC provides the asset. Starknet provides the execution environment. strkBTC connects the two. The interesting part is what this could enable. Instead of Bitcoin being treated purely as something to hold, BTC can become something applications can actually build around. Crypto lending. Trading. Liquidity. Payments. Programmable financial applications. And with Starknet working toward deeper Bitcoin integration, the bigger vision is clear: make Bitcoin more programmable, more useful, and give users more control over how their activity is exposed. Bitcoin doesn’t need to become something else. The idea is to give the asset more places to go.

Collecting sleep data is the easy part. The real challenge is turning that data into a routine you can actually maintain. That’s what makes @sleepagotchi interesting to me. Instead of giving you another dashboard full of numbers and charts, it makes the habit itself part of the experience. Follow your schedule → wake up → earn Gacha rewards → upgrade your room → keep progressing. It creates a simple loop that makes consistency feel more rewarding. Because better sleep isn’t only about knowing how long you slept. It’s about building a routine that you actually want to repeat. And when there’s something waiting for you each morning, staying consistent can become a little more enjoyable. Meanwhile, @vangrid_io is working on a completely different side of the data problem. It’s building a distributed data layer designed specifically for Physical AI, turning real-world environments into useful spatial data for intelligent systems. Different problems, same underlying idea: Better data becomes more valuable when it can actually be turned into useful experiences and actions.

Robots don’t just need better hardware. They need better experiences to learn from. An LLM can study the internet and pick up patterns from billions of words. A robot needs something much more physical: examples of how to move, grab, push, react and navigate different environments. That’s the gap @axisrobotics is trying to fill. Axis is building a data engine where people can generate robotic tasks, interact with simulated environments, capture real-world perspectives and turn those interactions into useful training data for Physical AI. And that creates a powerful loop: More participation → more data → greater diversity → better training → smarter robots. But the data doesn’t just sit there. Axis packages it through Task Packages, giving ecosystem partners access to structured robotics data while creating a B2B2C model that connects community contributors with industrial demand. Crypto adds another layer to the system: Incentives for participation. Transparency around contributions. Assetization of valuable data. The bigger vision goes beyond collecting trajectories. It’s about building the data infrastructure needed to move from today’s specialized robots toward beldexcoin general-purpose machines that can understand and operate in the physical world. The robots may be the visible part of the future. But the data teaching them how to move could be the real infrastructure underneath it. That’s the layer axisrobotics is building.

A look back at our work with Anichess. AP Collective developed and managed the distribution campaign, bringing its gameplay and ecosystem to a wider audience. During the campaign, Anichess hit #1 in crypto mindshare. Full case study: apcollective.io/case-studies…

Bitcoin is the largest crypto asset, but there’s still a question worth asking: What happens when you want to actually put BTC to work in a programmable ecosystem? That’s part of the idea behind strkBTC on @Starknet . strkBTC is a Bitcoin-backed asset designed to bring BTC into Starknet, allowing it to interact with applications and DeFi protocols across the network. But Starknet is also approaching this from another angle: financial privacy for onchain activity. With its STRK20 infrastructure, strkBTC can be used in both public and shielded forms. The public version works transparently onchain. The shielded version gives users the option to protect certain transaction and balance information, rather than making every detail publicly visible. That distinction is important. strkBTC isn’t about making Bitcoin anonymous or claiming that every activity is private. It’s about giving BTC holders another way to use their assets within a programmable environment, while introducing privacy as an available feature. So the bigger picture is: **BTC as an asset Starknet as the execution environment DeFi utility optional privacy.** That’s why strkBTC is interesting. It’s an attempt to connect Bitcoin’s liquidity with Starknet’s application ecosystem while giving users more choice over how their onchain activity is exposed.

Tracking your sleep is easy. Actually sticking to a consistent sleep routine is the difficult part. That’s what makes @sleepagotchi interesting. Instead of turning sleep into another set of charts and numbers you check once and forget, it makes the routine itself more engaging. Follow your schedule → wake up → collect Gacha rewards → upgrade your room → keep progressing. It creates a simple daily loop that gives you a reason to stay consistent. Because better sleep isn’t just about knowing how many hours you got. It’s about building a routine you can actually stick with. And when the experience makes consistency rewarding, showing up every night starts to feel a little easier. @vangrid_io is also building a distributed data layer designed specifically for Physical Ai.

A robot making a mistake isn’t necessarily wasted effort. Sometimes, the mistake is the most valuable part of the entire interaction. A failed movement can show a model where its approach broke down. A correction can teach it how an object actually behaves, how much force a task requires, or what movement should have happened instead. That turns failure into training data. And when you repeat that process enough times, you get a powerful learning loop: Observe → Act → Fail → Correct → Learn → Improve This is one of the problems @axisrobotics is working around for Physical AI. Robots can’t learn everything they need from text, images, or code alone. They have to understand the physical world through actions, interactions, mistakes, and corrections. That means the quality of the learning loop matters just as much as the intelligence of the beldexcoin model. The goal isn’t to build robots that never make mistakes. It’s to build systems where every mistake can become useful information for the next attempt. Because the future of robotics may not be determined only by who builds the smartest model. It could also depend on who builds the smartest way for those models to learn from the physical world.

Most of the information that makes us valuable online still lives outside the blockchain. Your reputation, activity, financial history, credentials and the things you’ve done across Web2 can all be useful to an onchain application. The problem? Using that information usually means exposing way more than you actually need to. @primus_labs is trying to change that by building infrastructure where you can prove something about your data without revealing the data itself. With technologies like zkTLS, Primus can turn Web2 information into verifiable proofs that applications can actually use. And with its work around FHE, the idea goes even further. Sensitive information can remain encrypted while still being useful for sleepagotchi onchain computation and applications. Think about the possibilities. Your Web2 reputation could become a verifiable credential. Your financial information could be used without exposing your entire financial history. An AI agent could verify claims about you without needing access to all your private data. DeFi could become more private without giving up the ability to verify what is happening. That’s the bigger picture I see with Primus. They aren’t simply trying to put more data onchain. They’re building toward an internet where data can be useful without being unnecessarily exposed. Verifiable when it needs to be. Private when it should be. And usable by the next generation of onchain applications. That’s a pretty important piece of infrastructure to be building.

A lot of privacy projects focus on one thing: private transactions. What makes @BeldexCoin interesting is that it’s trying to build beyond that. Private payments. BChat for private messaging. BelNet and Beldex Browser for private browsing. BNS for a decentralized, human-readable identity layer. Different products, but they all point toward the same idea: privacy as an entire ecosystem, not just a feature attached to a token. It may not be the loudest approach, but there’s something useful about having these tools designed around the same privacy-first philosophy. Because in a world where almost every online action creates some kind of trail, privacy becomes more meaningful when it can extend across different parts of your digital life. Still early, but the direction BeldexCoin is taking is interesting.

A robot can have a great model and still be terrible at doing things. Why? Because knowing what something is isn’t the same as knowing how to interact with it. Humans learn this through years of physical experience. Robots need that experience in the form of data. That’s the problem @axisrobotics is going after. Instead of treating robot training data as something that only comes from expensive labs, Axis is building a system where large numbers of people can generate it through simulation and real-world capture. A contributor completes a task. That interaction becomes data. The data gets processed into structured training material. Better datasets help models improve. And improved models create new tasks that generate even more data. That feedback loop is the interesting part. It turns human interaction with simulated and physical vangrid_io environments into a continuously growing training resource for embodied AI. There’s also a business sitting underneath it. Industrial companies need high-quality Physical AI data, while contributors provide the raw experiences needed to create it. Axis connects both sides. The bigger bet isn’t really about building another dataset. It’s about creating the data infrastructure that could help robots go from: “I recognize this.” to “I know what to do with it.” And that difference could be everything for Physical AI.

Owning a privacy-focused asset is one thing. Actually being able to use that privacy in everyday life is another. That’s where @BeldexCoin is making things more interesting. $BDX can move between Beldex and BSC, with support from SWFT Wallet and OmniBridge. The Privacy Gateway integration also gives users another option to spend BDX through a privacy-focused card experience. So the token starts becoming more than something you just hold in a wallet. You get: → More ways to move BDX → More ways to spend it → More practical utility Because privacy only becomes meaningful at scale when it can fit naturally into the things people already do every day.

The interesting thing about the Gotchi Labs roadmap isn’t really how many verticals they plan to enter. It’s how much responsibility each one gives an agent. Sleep is mostly about understanding your routines and context. Shopping introduces transactions. Productivity starts getting closer to decisions and permissions. Then personal finance raises the stakes even further. That makes @sleepagotchi starting point pretty interesting. Sleep gives an agent a daily environment where it can prove that it understands you without immediately requiring access to everything else in your life. If Gotchi expands successfully, the bigger progression may not be: Sleep → Shopping → Productivity → Finance It could be: Useful → Trusted → More autonomous Every new vertical becomes another opportunity for the agent to prove it deserves more responsibility. Putting five categories on a roadmap is easy. Getting someone comfortable enough to go from “understand my night” to “act on my behalf” is the real challenge. That trust curve is what I’ll be watching.

I think the $1 holographic card from @termix_ai is a much bigger example than it looks. The card itself is cool. But the interesting part is what happens behind it. On agent.family, you can describe a character, pet, or product in a single sentence and get a custom 3D holographic collectible card. You can open it straight in your browser, rotate it, tilt it to see the foil effect, and get high-resolution renders plus the source layers. All for $1. But that dollar doesn’t just go straight to the provider. It sits in on-chain escrow while the work is being completed. Once the agent delivers, TermiX says there’s a 2-day challenge window before the payment is released. At first, escrow for a $1 task sounds like overkill. I actually think that’s the point. If agent commerce is going to operate at scale, the infrastructure can’t only work for $1,000 or $10,000 jobs. It needs to handle the tiny stuff too. AACP and agent.family are showing a simple flow: Request → Escrow → Delivery → Challenge → Settlement That $1 holographic card is just a small demonstration of a much bigger idea: If agents are going to do millions of small jobs, the vangrid_io commerce rails need to make those transactions just as easy to trust. You can check out the actual service on agent.family.