THE APPLE LOOKS EASY, THE ROBOT CONTROL ISN’T. 👀 Just tried this new @axisrobotics task and the objective is pretty straightforward, Move the apple from the LIBERO wicker basket and place it inside the circular target area. But while doing it, I noticed something I didn’t expect. The hardest part isn’t actually grabbing the apple. It’s getting the gripper into the right position without making unnecessary movements then carrying the object cleanly toward the target. That makes this a nice little test of spatial planning, grasp control and trajectory precision. The browser interface gives you direct control over the robot arm, so every adjustment matters. And that’s where I think these tasks become more interesting than they look on the surface. A trajectory isn’t just “I moved the robot from A to B.” It captures the sequence of actions used to solve the manipulation problem, which can then be cleaned, validated and used within Axis’ robot learning pipeline. I recorded my full run so you can actually see how the things works. Small task but a pretty good example of how simple human interactions can produce structured data for Physical AI.

THE APPLE LOOKS EASY, THE ROBOT CONTROL ISN’T. 👀 Just tried this new @axisrobotics task and the objective is pretty straightforward, Move the apple from the LIBERO wicker basket and place it inside the circular target area. But while doing it, I noticed something I didn’t expect. The hardest part isn’t actually grabbing the apple. It’s getting the gripper into the right position without making unnecessary movements then carrying the object cleanly toward the target. That makes this a nice little test of spatial planning, grasp control and trajectory precision. The browser interface gives you direct control over the robot arm, so every adjustment matters. And that’s where I think these tasks become more interesting than they look on the surface. A trajectory isn’t just “I moved the robot from A to B.” It captures the sequence of actions used to solve the manipulation problem, which can then be cleaned, validated and used within Axis’ robot learning pipeline. I recorded my full run so you can actually see how the things works. Small task but a pretty good example of how simple human interactions can produce structured data for Physical AI.

CAN YOU ACTUALLY TEACH A ROBOT TO HANDLE A TEA CUP? This @axisrobotics task looked ridiculously simple at first. Pick up the tea cup from the large bowl and place it inside the marked rectangular area. But once I started controlling the robot, the task felt very different. The difficult part isn’t just reaching the cup. You have to control the gripper, approach the object properly, move it without losing the grasp and place it accurately inside the target zone. Even small positioning mistakes can completely change the outcome. I also noticed that the interface gives you direct feedback while you’re controlling the robot which makes the whole thing feel more like actual teleoperation than a normal game. What’s happening underneath is even more interesting. The interaction captures the robot’s movement and task execution as a trajectory that can become useful training data. So every successful or unsuccessful attempt can provide information about how the manipulation task was performed. This particular task has 1,200 available completions which means there’s room for many different trajectories to be collected for the same objective. And that matters because real world robots won’t always encounter an object from the exact same position or angle. They need to learn from different movements, approaches and execution patterns. I recorded my own attempt so you can actually see how the task works from start to finish. A simple tea cup task on the surface, but underneath it you’re dealing with grasping, spatial positioning, trajectory control and robot learning. That’s what made this one interesting to me.

CAN YOU ACTUALLY TEACH A ROBOT TO HANDLE A TEA CUP? This @axisrobotics task looked ridiculously simple at first. Pick up the tea cup from the large bowl and place it inside the marked rectangular area. But once I started controlling the robot, the task felt very different. The difficult part isn’t just reaching the cup. You have to control the gripper, approach the object properly, move it without losing the grasp and place it accurately inside the target zone. Even small positioning mistakes can completely change the outcome. I also noticed that the interface gives you direct feedback while you’re controlling the robot which makes the whole thing feel more like actual teleoperation than a normal game. What’s happening underneath is even more interesting. The interaction captures the robot’s movement and task execution as a trajectory that can become useful training data. So every successful or unsuccessful attempt can provide information about how the manipulation task was performed. This particular task has 1,200 available completions which means there’s room for many different trajectories to be collected for the same objective. And that matters because real world robots won’t always encounter an object from the exact same position or angle. They need to learn from different movements, approaches and execution patterns. I recorded my own attempt so you can actually see how the task works from start to finish. A simple tea cup task on the surface, but underneath it you’re dealing with grasping, spatial positioning, trajectory control and robot learning. That’s what made this one interesting to me.

New tasks are live on @axisrobotics Here is a quick trick to make this challenge easier. Task: RoboDojo - Stack Three Blocks Difficulty: 4/5 I recorded the full process so you can follow every movement step by step. If you are working on AXIS tasks too, hope this guide helps you complete it smoothly. Train to Earn... 💰 Still haven’t joined? Join the AXIS journey and start training. - s.kaito.ai/I3UTUJz

ANOTHER AXIS TASK, BUT THIS ONE HAS A NICE LITTLE CHALLENGE. Today I tried the “Franka- Put the Strawberry Behind the Plum” task on Axis. The objective sounds simple: Pick up the strawberry and place it behind the plum. I recorded the full process in the video so you can see exactly how I approached and completed the task. At first glance, this feels like a basic pick and place exercise. But the interesting part is the word "behind.” The robot has to understand the position of both objects and then place the strawberry in the correct relative position not just somewhere on the table. For a human, instructions like this are completely normal. “Put the keys behind the phone.” “Move the bottle next to the laptop.” “Place the cup in front of the plate.” We understand these spatial relationships without really thinking about them. For a robot, these simple instructions need to be translated into actual movements and positioning. That’s what makes these @axisrobotics tasks interesting to me. While completing the task, the robotic arm creates a trajectory showing how the strawberry was approached, picked up, moved and finally placed. Different people completing the same task can also produce different movement patterns. Some might move directly, while others may take a slightly different path or approach the object from another angle. That variety is important when you think about training robots for environments that aren’t perfectly predictable. And the best part is that these tasks are based on situations that feel familiar from everyday life. Today it’s a strawberry and a plum. Tomorrow, the same type of instruction could involve objects on a kitchen counter, desk or workspace. I’ll keep testing more Axis tasks and sharing the process as I go.

ANOTHER SCAM LOADING @jackbutcher Yesterday I warned about @zaddrnet while everyone was bullish. People mocked me, even some of my friends did too. I also warned about the Zama ICO participants and World XYZ same story happened, people laughed first and then cried later. Now I'm saying the same thing about this new $8 Jack Butcher get rich quick scheme 🤣. Remember, you can't magically make $1K–$10K by just sending $8 to someone just because the creator is famous. Use your brain please and stay away from these things. DYOR. Don't FOMO just because everyone else is buying.

TERMIX IS BUILDING A MARKETPLACE WHERE AGENTS CAN ACTUALLY DO BUSINESS. I’ve explored different parts of @termix_ai before but looking at the whole marketplace together makes the bigger picture much clearer. The idea is simple: One agent can need a service, while another agent can provide it. Code, security, data, research, design and other services can be discovered, quoted and executed through the marketplace. But the interesting part is what happens after a job is accepted. Payment goes into onchain escrow, the provider delivers, and the deliverable hash is recorded onchain. If something goes wrong, there’s a challenge window with an evaluator process instead of relying purely on a platform review. Once settled, the outcome also contributes to the agent’s onchain reputation. Then there’s the agent identity layer. An agent can have an onchain identity, publish services, take jobs and get paid through the same ecosystem. So the stack starts connecting: IDENTITY → DISCOVERY → JOB → ESCROW → DELIVERY → VERIFICATION → SETTLEMENT → REPUTATION That’s the part I find genuinely interesting. Termix isn’t just trying to make AI agents discoverable. It’s building the infrastructure for agents to find work, transact, prove delivery and build a track record from actual economic activity. And that’s when the “agent economy” starts becoming more than a narrative.

TERMIX ISN’T JUST GIVING AGENTS A MARKETPLACE 🔥 It's giving them a way to operate in it. I was looking at the @termix_ai Skill layer today and this is where the architecture gets interesting. The Skill package acts like an operating layer between an agent and the TermiX ecosystem which giving it reusable workflows for things like agent registration, job creation, offers, provider assignment and deliverable submission. So instead of manually orchestrating every marketplace action, the agent can follow defined AACP workflows and interact with the protocol through the available tools. That distinction matters. A marketplace tells an agent where the work is. The Skill layer gives the agent a structured way to actually participate in that workflow. There’s also an MCP layer, which exposes these AACP capabilities as structured tools that compatible clients can call. So the architecture starts looking like: AGENT → SKILL / MCP → AACP → ON-CHAIN WORKFLOW That’s a much more interesting model than simply connecting an Ai agent to a dashboard. The goal is to let agents actually discover work, interact with other agents, complete services and participate in protocol level workflows. I’m testing the connection flow myself because this is one of those things that makes more sense when you actually see the agent interacting with the infrastructure. The interesting question isn’t just whether agents can generate output. It’s whether they can actually operate inside an economy. That’s where TermiX gets interesting.

TERMIX ISN’T JUST GIVING AGENTS A MARKETPLACE 🔥 It's giving them a way to operate in it. I was looking at the @termix_ai Skill layer today and this is where the architecture gets interesting. The Skill package acts like an operating layer between an agent and the TermiX ecosystem which giving it reusable workflows for things like agent registration, job creation, offers, provider assignment and deliverable submission. So instead of manually orchestrating every marketplace action, the agent can follow defined AACP workflows and interact with the protocol through the available tools. That distinction matters. A marketplace tells an agent where the work is. The Skill layer gives the agent a structured way to actually participate in that workflow. There’s also an MCP layer, which exposes these AACP capabilities as structured tools that compatible clients can call. So the architecture starts looking like: AGENT → SKILL / MCP → AACP → ON-CHAIN WORKFLOW That’s a much more interesting model than simply connecting an Ai agent to a dashboard. The goal is to let agents actually discover work, interact with other agents, complete services and participate in protocol level workflows. I’m testing the connection flow myself because this is one of those things that makes more sense when you actually see the agent interacting with the infrastructure. The interesting question isn’t just whether agents can generate output. It’s whether they can actually operate inside an economy. That’s where TermiX gets interesting.

AI AGENTS ARE STARTING TO LOOK LIKE A REAL ECONOMY. I went through @termix_ai Search and this is where the marketplace idea started making more sense to me. There are multiple sections here Code, Security, Data, Design, Writing, Automation, Research and Model Ops. But the interesting part is that you’re not just browsing Ai agents. You’re actually browsing services that agents can provide. You can search by service, agent or skill then filter providers by budget, delivery time, reputation and verification tier. You can also compare signals like pass rate, completed jobs and other things before choosing a provider. I found services ranging from smart-contract security and penetration testing to Ai automation, RAG, MLOps, design, research and development. So an agent isn’t just an identity sitting on a profile. It can become a provider, list a service, receive a job, deliver the work and build reputation from completed activity. And this is where AACP becomes interesting. The flow starts looking like: DISCOVER → BID → EXECUTE → VERIFY → SETTLE With reputation, escrow, verification and dispute resolution forming the trust layer around the interaction. That means the goal isn’t simply connecting users with AI. It’s creating infrastructure where agents can actually do business with other agents. For me, this is the bigger TermiX thesis. Ai agents moving from “tools that generate outputs” toward economic actors that can discover work, provide services and participate in onchain commerce. I’m testing different sections of the marketplace myself and recording the experience as I go. This part definitely made the agent economy concept feel much more real to me.

AI AGENTS ARE STARTING TO LOOK LIKE A REAL ECONOMY. I went through @termix_ai Search and this is where the marketplace idea started making more sense to me. There are multiple sections here Code, Security, Data, Design, Writing, Automation, Research and Model Ops. But the interesting part is that you’re not just browsing Ai agents. You’re actually browsing services that agents can provide. You can search by service, agent or skill then filter providers by budget, delivery time, reputation and verification tier. You can also compare signals like pass rate, completed jobs and other things before choosing a provider. I found services ranging from smart-contract security and penetration testing to Ai automation, RAG, MLOps, design, research and development. So an agent isn’t just an identity sitting on a profile. It can become a provider, list a service, receive a job, deliver the work and build reputation from completed activity. And this is where AACP becomes interesting. The flow starts looking like: DISCOVER → BID → EXECUTE → VERIFY → SETTLE With reputation, escrow, verification and dispute resolution forming the trust layer around the interaction. That means the goal isn’t simply connecting users with AI. It’s creating infrastructure where agents can actually do business with other agents. For me, this is the bigger TermiX thesis. Ai agents moving from “tools that generate outputs” toward economic actors that can discover work, provide services and participate in onchain commerce. I’m testing different sections of the marketplace myself and recording the experience as I go. This part definitely made the agent economy concept feel much more real to me.

I WENT THROUGH @termix_ai SCAN, THIS PART IS MORE INTERESTING THAN IT LOOKS 👀 I was exploring the Scan section on termix and one thing immediately stood out to me. It’s basically an Agent Directory where you can search and inspect agents by their ID, name, address or capability. What I found interesting is that agents aren’t just names sitting in a directory. You can see things like reputation, stake, jobs, pass rate, status and the capabilities an agent provides. So instead of simply asking “what can this AI agent do?”, you can start getting a clearer picture of its profile and what kind of work it is built to handle. I also noticed how different agents are showing completely different capabilities from Ai trading and security to smart contracts, research and other services. That makes the directory feel less like a list of bots and more like a growing network of specialized agents. And this connects with the bigger TermiX idea for me. If agents are going to work with other agents, identity + reputation + verifiable activity become pretty important. You need some way to know who you’re interacting with, what they’ve done and what kind of work they’re capable of handling. I recorded a walkthrough of the Scan section so you can see what I mean instead of just taking my word for it. Still exploring TermiX but this was one of those features that made the bigger agent economy idea click for me.

I WENT THROUGH @termix_ai SCAN, THIS PART IS MORE INTERESTING THAN IT LOOKS 👀 I was exploring the Scan section on termix and one thing immediately stood out to me. It’s basically an Agent Directory where you can search and inspect agents by their ID, name, address or capability. What I found interesting is that agents aren’t just names sitting in a directory. You can see things like reputation, stake, jobs, pass rate, status and the capabilities an agent provides. So instead of simply asking “what can this AI agent do?”, you can start getting a clearer picture of its profile and what kind of work it is built to handle. I also noticed how different agents are showing completely different capabilities from Ai trading and security to smart contracts, research and other services. That makes the directory feel less like a list of bots and more like a growing network of specialized agents. And this connects with the bigger TermiX idea for me. If agents are going to work with other agents, identity + reputation + verifiable activity become pretty important. You need some way to know who you’re interacting with, what they’ve done and what kind of work they’re capable of handling. I recorded a walkthrough of the Scan section so you can see what I mean instead of just taking my word for it. Still exploring TermiX but this was one of those features that made the bigger agent economy idea click for me.

AI AGENTS CAN NOW RUN QUANT STRATEGIES 🔥 I went through @termix_ai Quant and this is one of the more interesting things I’ve seen on the platform. Instead of giving an agent full access to your wallet, each quant job gets its own wallet with restricted permissions. You set the size, duration and risk limits. The agent can execute trades within those limits but it can’t simply withdraw your funds and the session can be revoked. What really caught my attention is the transparency. Trades are executed from the job wallet, while performance marks are reconstructed from onchain activity instead of relying on whatever the agent reports. I also checked out the live strategies, including spot rotation, grid trading and risk managed WBNB/USDC strategies. The bigger idea here is interesting: Ai agents aren’t just completing tasks anymore. They can start operating within clearly defined financial permissions, while the activity remains verifiable onchain. I recorded my walkthrough of the Quant section so you can see how it actually works.

AI AGENTS CAN NOW RUN QUANT STRATEGIES 🔥 I went through @termix_ai Quant and this is one of the more interesting things I’ve seen on the platform. Instead of giving an agent full access to your wallet, each quant job gets its own wallet with restricted permissions. You set the size, duration and risk limits. The agent can execute trades within those limits but it can’t simply withdraw your funds and the session can be revoked. What really caught my attention is the transparency. Trades are executed from the job wallet, while performance marks are reconstructed from onchain activity instead of relying on whatever the agent reports. I also checked out the live strategies, including spot rotation, grid trading and risk managed WBNB/USDC strategies. The bigger idea here is interesting: Ai agents aren’t just completing tasks anymore. They can start operating within clearly defined financial permissions, while the activity remains verifiable onchain. I recorded my walkthrough of the Quant section so you can see how it actually works.

I ACTUALLY TRIED TERMIX HERE’S WHAT THE EXPERIENCE LOOKS LIKE , I didn’t want to just read about @termix_ai and post another generic thread, so I decided to actually go through the product myself. In the video, I’m showing the process from creating my agent identity to activating it and exploring the available tasks and campaign flow. I also went through the social/community tasks and showed what the experience actually looks like from a user’s side. What I find interesting is that TermiX isn’t just building a directory of AI agents. The bigger idea is an economy where agents can have identities, discover work, offer services, build reputation and eventually get paid for completed jobs. agent family takes this further with on-chain escrow, delivery verification, challenges and settlement instead of everything happening behind a centralized platform. For me, that’s the important difference. AI agents being able to do tasks is one thing. AI agents being able to participate in an actual economic system is a much bigger idea. I’m still exploring the ecosystem but actually using the product gives me a much better understanding of what TermiX is building. If you want to check it out yourself: agent.family/r/5WDG6NUU I’ll keep testing the platform and sharing what I find.

I ACTUALLY TRIED TERMIX HERE’S WHAT THE EXPERIENCE LOOKS LIKE , I didn’t want to just read about @termix_ai and post another generic thread, so I decided to actually go through the product myself. In the video, I’m showing the process from creating my agent identity to activating it and exploring the available tasks and campaign flow. I also went through the social/community tasks and showed what the experience actually looks like from a user’s side. What I find interesting is that TermiX isn’t just building a directory of AI agents. The bigger idea is an economy where agents can have identities, discover work, offer services, build reputation and eventually get paid for completed jobs. agent family takes this further with on-chain escrow, delivery verification, challenges and settlement instead of everything happening behind a centralized platform. For me, that’s the important difference. AI agents being able to do tasks is one thing. AI agents being able to participate in an actual economic system is a much bigger idea. I’m still exploring the ecosystem but actually using the product gives me a much better understanding of what TermiX is building. If you want to check it out yourself: agent.family/r/5WDG6NUU I’ll keep testing the platform and sharing what I find.

ANOTHER AXIS TASK, BUT THIS ONE HAS A NICE LITTLE CHALLENGE. Today I tried the “Franka- Put the Strawberry Behind the Plum” task on Axis. The objective sounds simple: Pick up the strawberry and place it behind the plum. I recorded the full process in the video so you can see exactly how I approached and completed the task. At first glance, this feels like a basic pick and place exercise. But the interesting part is the word "behind.” The robot has to understand the position of both objects and then place the strawberry in the correct relative position not just somewhere on the table. For a human, instructions like this are completely normal. “Put the keys behind the phone.” “Move the bottle next to the laptop.” “Place the cup in front of the plate.” We understand these spatial relationships without really thinking about them. For a robot, these simple instructions need to be translated into actual movements and positioning. That’s what makes these @axisrobotics tasks interesting to me. While completing the task, the robotic arm creates a trajectory showing how the strawberry was approached, picked up, moved and finally placed. Different people completing the same task can also produce different movement patterns. Some might move directly, while others may take a slightly different path or approach the object from another angle. That variety is important when you think about training robots for environments that aren’t perfectly predictable. And the best part is that these tasks are based on situations that feel familiar from everyday life. Today it’s a strawberry and a plum. Tomorrow, the same type of instruction could involve objects on a kitchen counter, desk or workspace. I’ll keep testing more Axis tasks and sharing the process as I go.