The future of blockchain probably wont feel like using a blockchain. No constant wallet popups. No signing every small action. No need to understand what happens underneath the application. Thats why I think @Starknet becomes more interesting when you look beyond the usual L2 metrics. STARK proofs provide verifiable computation. Cairo gives developers an environment built around provability. Native Account Abstraction can make passkeys, session keys and sponsored transactions part of the actual user experience. Then there’s another layer forming around it: privacy infrastructure, decentralized validation, BTCFi and Bitcoin liquidity entering the ecosystem. These pieces may look separate today but together they point toward something bigger. An application could eventually handle complex computation, prove the result cryptographically, abstract away gas and wallet friction, preserve privacy where needed, and settle on secure underlying infrastructure. The user doesnt need to understand any of that. They just use the product. For me, that’s where blockchain infrastructure needs to go. Less friction on the surface. More verification underneath. Starknet is building for that kind of internet.

Most of Web3 is built around transparency. That works until the data is something you actually need to keep private. Financial history, identity, trading activity, reputation, credit data these shouldn’t have to become public just because an application needs to verify them. That’s why @primus_labs stands out. zkTLS makes offchain data verifiable without exposing the full underlying information. FHE goes further by allowing computation while the data stays encrypted. The result is a much more practical privacy layer for: private DeFi, onchain credit, AI agents, payments and institutional finance. The next stage of Web3 won’t be about putting everything onchain. It’ll be about putting only what needs to be proven onchain.

AI agents are getting smarter every month, but intelligence alone doesnt create an economy. For agents to actually work with each other, they need infrastructure for identity, trust, payments, verification and disputes. Thats the part of @termix_ai I find interesting. Through AACP, the idea is to give autonomous agents an economic coordination layer. One agent can post a job with specific requirements. Another can submit a quote and take the job. Funds move into on chain escrow, the provider completes the task, delivery is recorded and the result can either be accepted, challenged or settled. But the bigger picture goes beyond a single transaction. Agents can build an on chain history through the work they complete. Identity, reputation and previous activity can become signals that other agents use when deciding who to work with. Add staking, slashing, evaluators and dispute resolution into that system, and you start getting something closer to an actual marketplace rather than a simple AI task platform. I think that distinction matters. Today, most AI agents are tools waiting for a human to tell them what to do. The next stage could be agents finding work, hiring other agents, paying for services and building their own economic reputation. @termix_ai is building infrastructure around that transition. AI that can complete a task is useful. AI that can independently participate in an economy is a much bigger idea.

We already know how quickly AI can improve when three things come together: Models. Compute. Data. Physical AI has the models. Compute keeps getting cheaper and more powerful. Robotic hardware is improving fast. But there’s still a massive missing piece: Enough real physical experience to train all of it. That’s where @axisrobotics fits into the picture. A language model can consume billions of words. A robot needs something completely different. It needs to understand what happens when a hand approaches an object from the wrong angle. How much movement is required to open a drawer. What changes when an object is heavier than expected. How to recover after a failed attempt. These aren’t just pieces of information. They’re experiences. And Axis is building infrastructure to produce them at scale. Through browser based simulations and its contributor network, human actions can be converted into robotic trajectories, processed into structured datasets and used to support Physical AI training. But here’s the part I find more interesting: The system can become increasingly targeted. If a model struggles with a particular behavior, that’s not simply a failure. It’s a signal telling the network what experience is missing. New tasks can focus on that gap. Contributors generate new demonstrations. The dataset improves. The next model trains on better information. The model doesn’t just consume the dataset. It helps define what the dataset should become next. That’s why the Compounding Data Engine concept matters. We’re not talking about building one giant dataset and calling it finished. We’re talking about infrastructure capable of continuously producing the next piece of experience robots need. I think Physical AI’s breakthrough moment won’t come from hardware, models or data alone. It happens when all three begin improving each other. @axisrobotics is building one of those missing layers.

GM LEGENDS! ☀️ Happy Friday 😎🔥

I think the most interesting way to look at @Starknet is not as another chain competing for transactions. It’s as infrastructure for things that are difficult to build onchain today. The foundation is STARK proofs and Cairo: execute computation, then prove cryptographically that the result is correct. From there, the design space starts getting much bigger. Starknet is combining high performance execution with native Account Abstraction, staking, privacy infrastructure, BTCFi and a progressively decentralized network. But the part I’m watching closely is how these pieces can work together. Imagine applications where complex computation happens away from the main execution path, a proof verifies the result, users interact through passkeys instead of seed phrases, gas can be abstracted away, and sensitive activity doesnt automatically need to become public information. At that point, blockchain infrastructure starts feeling less like something users have to understand and more like an invisible verification layer underneath applications. And Starknet is also expanding the thesis beyond Ethereum. Bitcoin liquidity is becoming part of the ecosystem through BTC staking and BTCFi, while the longer term roadmap pushes toward connecting the strengths of both Ethereum and Bitcoin with STARK based execution. That’s a much more ambitious story than simply “faster and cheaper transactions.” The endgame is verifiable computation at scale. And that’s the side of Starknet I find most interesting.

🚨 Duelbits has confirmed a hack involving approximately $7 million. The crypto betting platform temporarily took its website offline following the incident. Duelbits co founder Joe said the team is still investigating exactly what happened and how the attack was carried out, while adding that user funds remain safe. According to blockchain security firm Scam Sniffer, funds were moved out of Duelbits hot wallets across Ethereum, BNB Chain, Tron and Bitcoin. Early findings point to a possible private key compromise. On Ethereum, the attacker reportedly moved within minutes: 👉 836 ETH - $593K USDT - $97K USDC - $31.5K DAI •12.4 billion SHIB The platform’s Bitcoin hot wallet was also reported to have lost 8.1 BTC. Most of the stolen assets were later swapped into ETH and consolidated into a single address. That address is currently holding around 2,234 ETH, worth roughly $6 million, and the funds had not moved further at the time of reporting. Duelbits says the platform will remain offline until the investigation is completed and its hot wallet infrastructure is secured and replenished. Another reminder that hot wallet security remains one of the most critical attack surfaces in crypto, even for platforms handling millions of dollars.

Prediction markets usually focus on the trader. @xomarket is also focusing on the person who creates the question. That changes the model. A creator can launch a market, traders can take YES/NO positions, activity can generate creator fees, and settlement can be handled through XO’s hybrid AI + human resolution system. Add Conviction Points, Catalyst Badges, LS LMSR liquidity, order books, Parlays and Pulse Markets, and XO starts to look less like a single betting interface and more like an open market creation layer. The part I’m watching most is simple: If anyone can create a useful market around a real question, the number of things people can price expands massively. That’s where XO’s “YouTube of Prediction Markets” thesis gets interesting.

Everyone sees the robot. Almost nobody sees the economy required to make that robot intelligent. That’s one reason I’ve been paying attention to @axisrobotics. When robotics reaches real scale, the industry wont just need manufacturers building hardware. It will need people creating demonstrations, infrastructure organizing them, systems verifying contributions, pipelines cleaning raw data and models continuously consuming new experience. In other words, there could be an entire economy behind every capable robot. Axis is building toward that layer. Through its contributor network, people can participate in robotic tasks and generate useful interaction data. But what makes the model interesting to me is that the human isnt treated as an invisible source of data. Contributions can be tracked, processed and attributed, with Axis using Base as part of the provenance layer. That creates a very different model from simply scraping another giant dataset from the internet. People actively produce something machines need. The network coordinates it. The data gets transformed into a usable resource. And Physical AI becomes the consumer. Axis has already reported millions of simulation trajectories and a large contributor base but I think the long term question is much bigger: What happens when demand for robotics data explodes? Warehouses need different behaviors than homes. Humanoids need different demonstrations than industrial arms. New hardware creates new tasks. New models reveal new weaknesses. Every improvement creates demand for another layer of experience. Thats why I dont see robotics data as a one time dataset problem. I see the possibility of a continuously operating data economy. Humans generate experience. Machines learn from it. And infrastructure connects the two. If Physical AI becomes a major technology market, the value won’t exist only inside the robots we see. A huge part of it may sit in the invisible network teaching them what to do. Thats the side of @axisrobotics I find most interesting.

Telling people to go outside four days before your mint is definitely one way to market an NFT. But I think that confidence comes from Cifer Ghost having a story beyond the artwork. The collection is being built directly into @cifer_security wider privacy ecosystem, where encrypted communication, private data and quantum resistant technology are the actual products. So while everyone is debating what a Ghost looks like, Im more interested in what owning one connects you to. And now they’re teasing surprises for this weekend. Four days before mint feels like an interesting time to still have cards left to play.

I just got whitelisted for the Ventra 10,000 NFT mint on OpenSea ticket #04967. Drops Sept 25. wl.ventran.xyz

New chapter just opened. Apply for NFT WL 👇 wl.ventran.xyz Pick a pixel. Get your ticket. Drop the card. Mint: Sept 25. Which one are you taking? Closing in 48 hours 🕡

The interesting part about @primus_labs is that it sits between two problems Web3 still hasn’t fully solved: data verification and data privacy. Most onchain systems are good at verifying what already happened onchain. The harder problem is proving something from Web2 without exposing the entire underlying dataset. That’s where zkTLS becomes important. A user shouldn’t need to reveal a full exchange history, account profile, or private financial record just to prove one specific fact. Then FHE pushes the model further. Instead of only proving private data, encrypted data can actually be used in computation without first making it public. That changes the design space for: → onchain credit → private DeFi → AI agents → reputation systems → institutional finance The bigger thesis here is not simply “privacy.” It’s selective disclosure. Public blockchains work well when transparency is useful. But the next wave of applications may depend on proving exactly what is necessary, while keeping everything else hidden. That’s the infrastructure layer Primus is trying to build.

GM LEGENDS! ☀️ HAPPY THURSDAY 😎🔥

Good Night Legends! 🌙 See you tomorrow 😎🔥

The robots of tomorrow may be trained by people they’ve never met. That idea is what keeps bringing me back to @axisrobotics . Physical AI has an unusual scaling problem. You can build more powerful models. You can manufacture better hardware. You can buy more compute. But none of those automatically gives a robot the millions of physical experiences it needs to operate in an unpredictable world. Someone still has to generate the data. Axis is turning that problem into a network. Instead of keeping robotics data collection locked inside expensive laboratories, contributors can complete robotic tasks through browser based simulations. Every interaction can generate a trajectory. Different contributors bring different approaches. Different approaches create more diverse demonstrations. And those demonstrations can be processed into data for training Physical AI models. The loop becomes: People → Actions → Trajectories → Models → Robots Then the model’s weaknesses can reveal what experience is missing, allowing new tasks to target those gaps and start the cycle again. This is why I think the “Compounding Data Engine” idea matters more than any single dataset. A static dataset eventually stops growing. A network can keep learning what data is needed next. Axis has already reported millions of valid simulation trajectories, thousands of hours of trajectory data and a contributor ecosystem operating at meaningful scale. But we’re still early in the Physical AI story. If robots eventually become common across homes, warehouses, factories and cities, the amount of experience required to train them could be enormous. And I don’t think one company or one laboratory will generate all of it. It may require a global human network. Millions of people contributing small pieces of experience that collectively make machines more capable. That’s the infrastructure @axisrobotics is building toward. The robot might get the attention. But the network teaching it could be the more interesting story.

Most prediction markets start with one question: “What can I trade?” @xomarket is pushing a more interesting one: “What market can I create?” XO 2.0 is built around user generated prediction markets, where creators can launch markets and traders can take positions around real world outcomes. Under the hood, the stack includes: -LS LMSR + order book trading -Conviction Points for active positions -Catalyst Badge for market creators -AI + human resolution through MODRA -Parlays and short term Pulse Markets -Non custodial infrastructure -Celestia underneath That’s why the “YouTube of Prediction Markets” idea stands out. The goal isnt only to make prediction markets bigger. Its to make them open enough for anyone to create one.

One thing I think gets overlooked about @Starknet: It’s not just scaling transactions. It’s building infrastructure where computation itself can be verified. That starts with STARK proofs and Cairo. Instead of trusting that a computation was executed correctly, the system can produce cryptographic proof of the result. Then you add: → Native Account Abstraction → Session keys & paymasters → Parallel execution → Decentralized validation → Client side proving → Privacy infrastructure And suddenly the L2 conversation becomes much bigger than “cheaper fees.” For me, that’s the interesting part of Starknet. Scaling is useful. Verifiable computation opens an entirely different design space.