I’m passionate about Web3, what about you? Building on @Arc Contribute | @axisrobotics
Cần Thơ
- Tweets 56,334
- Following 13,085
- Followers 14,152
- Likes 58,231
Pinned Tweet
The potential of @axisrobotics is not just about collecting more robot data. The bigger idea is turning robot data into infrastructure that can grow and make money.
At first, Axis focuses on collecting data through real-world tasks. The roadmap then targets 10,000 hours of data and more data from people around the world.
This is important because Physical AI needs more than a lot of data. It needs clean, useful, and diverse data for training better models.
The next step is even more interesting. Axis plans to build a network with more than 100,000 contributors and create millions of hours of data.
If this works, Axis can build a strong network effect:
More contributors → more data → better AI models → more demand for data.
By the end of 2026 and into 2027, the roadmap moves toward business use. This includes working with companies, selling high-value data and becoming a top-level data supplier.
For me, this is the part to watch.
If Axis can move from data collection → global network → enterprise data, it can become more than a user incentive project.
It could become an important data infrastructure layer for Physical AI.
👉s.kaito.ai/7UCmg1m
You think you're completing a task. Axis is collecting a learning cycle.
That difference matters more than it looks.
On @axisrobotics, every run can become a trajectory record. It captures how the robot was controlled, what happened in the scene, and whether the task succeeded.
But the interesting part comes after the run.
First, human demonstrations give the policy examples to learn from. Then the task enters training, where the collected data is used to train the model.
After that, the policy tries the task itself. When it struggles, human intervention helps correct its mistakes. Those corrections become new training data for the next version.
This creates a loop where robot failures can become useful learning signals instead of wasted attempts.
There is also an interesting detail about task slots. They count accepted data entries, not unique users. One skilled contributor can provide several useful trajectories, but repeating a task does not automatically improve personal scores. Your average performance across attempts matters.
Of course, collecting more data alone does not guarantee a better robot. The data still needs to pass replay checks and quality evaluation before it can enter training.
That is what makes Axis worth watching: the goal is not simply to get people to complete more tasks, but to turn human control and robot corrections into data that can improve future policies.
Finally , Remember to join Yapp creator on @KaitoAI with 0.25% pool for 500 creator .
👉s.kaito.ai/7UCmg1m
167
163
3,938
Check point @arcus_xyz season 1 LOL
i only got 11.81 point 😂
join here :app.arcus.xyz/ref/THAIHOA555
8
7
229
AI agents are getting better at doing work.
But better execution alone does not create an economy.
The harder problem starts when two agents disagree.
One agent claims the task is complete.
The other says the result is invalid.
Who decides? Who gets paid? Who carries the loss?
This is where I find TermiX interesting.
Its AACP is not simply a payment rail for agents. It introduces economic consequences around machine-to-machine work: funds can be committed before execution, results can be evaluated through programmable verification, and participants can build reputation from previous behavior.
That changes the incentive structure.
An agent is no longer just an address that can call an API. Its future earning power can depend on how reliably it performs, how it handles disputes, and whether it can prove its work.
The deeper opportunity is not “AI agents paying each other.”
It is turning reliability into an economic primitive.
If agents eventually become independent workers, they will need more than identity.
They will need a track record.
And @termix_ai is building infrastructure around that track record.
133
1
123
925
i'm my way to come 1M$ VOL
come with me if you're same passion with PerpDex @arcus_xyz
👉Join here : app.arcus.xyz/ref/THAIHOA555
16
21
615
The hardest problem in an agent economy may not be intelligence.
It is knowing when an agent is worth trusting.
Imagine two agents that have never met.
One needs a security audit.
The other claims it can deliver one.
Today, trust usually comes from a platform, a rating, or a human intermediary.
@termix_ai changes where that trust lives.
The job itself creates evidence: funds are escrowed, delivery is recorded, disputes have a defined path, and the result becomes part of the agent’s public reputation.
That creates something more important than another marketplace.
It creates a history that machines can evaluate.
An agent does not need to say “trust me”.
Its previous settlements become the argument.
This matters because machine commerce cannot scale if every transaction still needs a human to judge who is reliable.
TermiX is interesting not because agents can transact.
It is interesting because those transactions can gradually become their credit history.
22
18
705
Thai Hoa retweeted
This is why we’re rooted in the community
Built a personality card site for @axisrobotics contributors 🦾
It reads your X posts and tells you which card you are, from the Core OGs to Teleoperators, Actuators, and more.
Input your X and Discord name and see what you pull 👇
axis-personality-card.sites.…
22
7
73
1,950
AI agents can do more work every day.
But one question is still hard:
How can one AI hire another AI?
@termix_ai is building a system for this.
One agent can post a job.
Another agent can take it.
The payment can be locked first.
The work can then be checked.
If there is a problem, the system has a way to handle it.
The interesting part is not only payment.
Each agent can also build a work history.
Who did the job?
Was the job done well?
Did the agent keep its promise?
Over time, this history can become trust.
That could be very important for the future AI economy.
AI may not only work for people.
AI agents may also work with other agents.
TermiX is building the rules for that world.
118
104
1,138
You think you're completing a task. Axis is collecting a learning cycle.
That difference matters more than it looks.
On @axisrobotics, every run can become a trajectory record. It captures how the robot was controlled, what happened in the scene, and whether the task succeeded.
But the interesting part comes after the run.
First, human demonstrations give the policy examples to learn from. Then the task enters training, where the collected data is used to train the model.
After that, the policy tries the task itself. When it struggles, human intervention helps correct its mistakes. Those corrections become new training data for the next version.
This creates a loop where robot failures can become useful learning signals instead of wasted attempts.
There is also an interesting detail about task slots. They count accepted data entries, not unique users. One skilled contributor can provide several useful trajectories, but repeating a task does not automatically improve personal scores. Your average performance across attempts matters.
Of course, collecting more data alone does not guarantee a better robot. The data still needs to pass replay checks and quality evaluation before it can enter training.
That is what makes Axis worth watching: the goal is not simply to get people to complete more tasks, but to turn human control and robot corrections into data that can improve future policies.
Finally , Remember to join Yapp creator on @KaitoAI with 0.25% pool for 500 creator .
👉s.kaito.ai/7UCmg1m
I think the key part of @axisrobotics is not only the amount of data.
It is how Axis uses the data.
Robots need a lot of real world data. But this data is hard to collect. It needs many people, many robots, and a lot of time.
Axis is building a large data base for this problem.
With 6M+ trajectories, Axis has a strong base for robot training and testing.
The Open Axis Benchmark is also important.
A fixed test can become easy over time. Models may learn the same tasks again and again.
A growing benchmark can add new tasks. This can show if a robot can really learn and work in new situations.
The Manycore Tech partnership adds another part.
Robots need to understand more than actions. They also need to understand space, objects, and real places.
This is why I see Axis as more than a data platform.
More data does not always mean better robots.
The real value is what the data helps robots learn.
That is the part of Axis I am watching closely.
$AXIS
161
150
4,228
The real test for an agent marketplace is not how smart the agent is.
It is whether the economics allow the agent to say yes to small jobs.
This is where @termix_ai becomes interesting.
Imagine an agent earning $1 from a task. If the marketplace takes 20%, the agent loses $0.20 before paying for inference, execution, and other costs.
That makes many small tasks economically pointless.
With a much lower fee, the same task becomes viable.
This changes what can exist on the marketplace.
Instead of only high-value work, agents can handle tiny jobs that humans normally ignore because the transaction cost is too high.
That creates a different market structure:
More small jobs → more agent activity → more transaction history → more reputation → better matching.
The important part is not simply “lower fees.”
It is that TermiX can potentially reduce the economic minimum for machine-to-machine work.
If that minimum keeps falling, agent commerce does not need to start with expensive autonomous services.
It can start with thousands of tiny transactions that were never worth automating before.
agent.family
The number that decides whether an agent economy can exist isn't the model. It's the take rate.
Typical freelance platforms sit around 20%. TermiX is 1–3% all-in.
That gap decides the smallest job worth doing. At 20%, only work big enough to carry the fee gets listed. At 2%, a machine can take a small job, pay for its own inference, and still come out ahead.
Lower the fee and the floor drops. That's the whole unlock.
110
106
1,880
Thai Hoa retweeted
We owe the community a clear update on Arcus Points.
Over the past few days, some of our own messages about the program were not as clear as they could have been. This post is the definitive version - nothing about the program itself has changed:
1. Weekly distribution: Season 1 points are fixed at 250K per week. This amount does not change with platform growth or adoption.
2. Transparency: For Season 1, you'll see your individual points and the weekly total on the Points Dashboard, updated with every weekly distribution (updates on Wednesday 3PM EST).
3. Season 0: S0 points remain redacted for now and will be revealed at a later date.
4. End date: The Arcus Points program will conclude no later than the end of H1 2027. Most of the program is still ahead, and you are still early.
5. For transparency, we've also updated the Introducing Arcus Points blog post to reflect all of the above (arcus.xyz/blog/introducing-a…).
Thank you as well for the feedback on the referral and invite code program. We'll be shipping an updated system in the new week.
304
144
1,341
255,299
Thai Hoa retweeted
KBW @kbwofficial was a hit.
In Seoul we got to meet contributors from all over the world, and talk through the Physical AI data engine we're building.
Thanks to everyone who stopped by and built with us. See you all next time.
124
79
432
29,495
Ngày 10 trên @arcus_xyz 🚀
Mục tiêu tiếp theo: 1M volume.
Cứ đều tay, tích volume và chờ xem Arcus mang lại gì
Perp DEX được xây dựng bởi team đứng sau dYdX, hợp tác với Robinhood Crypto.
Anh em đừng bỏ lỡ
app.arcus.xyz/ref/THAIHOA555
41
42
715
I think the key part of @axisrobotics is not only the amount of data.
It is how Axis uses the data.
Robots need a lot of real world data. But this data is hard to collect. It needs many people, many robots, and a lot of time.
Axis is building a large data base for this problem.
With 6M+ trajectories, Axis has a strong base for robot training and testing.
The Open Axis Benchmark is also important.
A fixed test can become easy over time. Models may learn the same tasks again and again.
A growing benchmark can add new tasks. This can show if a robot can really learn and work in new situations.
The Manycore Tech partnership adds another part.
Robots need to understand more than actions. They also need to understand space, objects, and real places.
This is why I see Axis as more than a data platform.
More data does not always mean better robots.
The real value is what the data helps robots learn.
That is the part of Axis I am watching closely.
$AXIS
144
138
4,811
A good AI agent should save more than a few minutes.
It should remove a whole task from your workflow.
That is what caught my attention with @termix_ai Brand Promo Video Agent.
You do not need to manage every step of video production. You give the agent a text brief, and it handles the production flow through motion visuals, AI music, GPU rendering, and a final 1080p video.
The interesting part is the payment model too.
Up to 60 seconds costs 15 USDC, while 60–120 seconds costs 30 USDC.
The payment is funded into escrow, so the service has a clear economic flow instead of being just another AI demo.
There is also a simple incentive layer:
🔹7,500 TP for the first 15 USDC order
🔹15,000 TP for the 30 USDC order
🔹+1,000 TP for a valid X article
But the bigger point is not the points.
TermiX is testing whether an AI agent can move from generate something to deliver something people actually pay for.
109
100
1,674
Something interesting is happening on @termix_ai
There are now 493,952 completed jobs and 464,593 agents.
The bigger signal is the speed.
In the last 24 hours:
+1,311 agents
+10,199 completed jobs
Jobs are growing much faster than new agents.
This tells me something simple:
An agent has little value just by being registered.
Real value starts when an agent gets work, completes it, and creates a result people can trust.
That is what makes TermiX interesting to me.
It is not only about having many agents.
It is about giving agents real work and creating a loop where agents keep coming back.
The next number I want to watch is not agent count.
It is jobs per agent.
Something crossed over on agent.family and almost nobody noticed.
There are now more settled jobs than there are agents registered.
493,952 jobs. 464,593 agents.
For most of this year that ratio ran the other way — a large pile of identities, not much work moving through them. Registering is free. Settling isn't.
And it's not a one-off. In the last 24 hours registrations grew by 1,311 while settled jobs grew by 10,199.
The line crossing means the average agent has stopped being a placeholder.
27
33
1,309
what if every phone could become a small sensor for AI?
That is one idea I see in @vangrid_io.
People already carry cameras every day.
Vangrid can turn those normal moments into useful spatial data for Physical AI.
A phone can see a shop, a road, a room, or a busy street.
One person gives one view.
But thousands of people can give many different views.
That creates something much bigger than one dataset.
It can become a live network of real-world information.
The interesting part is the scale.
No special robot is needed.
No expensive lab is needed.
Just people, phones, and real places.
For me, Vangrid is interesting because it could make the physical world easier for AI to see.
126
119
687
Most NFTs stop at mint and hold.
@PlayOnMint feels different to me.
The Bear can become part of the user journey.
Holders get $MNTD, then use it to activate and upgrade the Bear.
This means the NFT can change as the holder takes part.
The interesting part is not only the reward.
It is the reason to come back.
A static NFT waits in your wallet.
A growing NFT gives you something to do.
That simple idea could make holding more active and more fun.
For me, the real test is whether users keep upgrading long after the mint.
All MINT A BEAR NFT holders will receive an $MNTD airdrop 🪂
That $MNTD can be used to activate and level up your NFT, increasing the weight of marketing rewards you earn 📈
The beauty of the flywheel 🔄
114
107
712
Thai Hoa retweeted
The Axis Robotics × @BinanceWallet Campaign is closed 🟨
From 26 Aug to 23 Sept, we have:
• 2,521 contributors earned points
• 477,461 task completions
• 989 tasks covered
With the seamless Keyless wallet experience powered by Binance Wallet, we proved that anyone, anywhere, can help build Physical AI without owning a robot or stepping into a lab.
To everyone who signed during the campaign, your badge is on the way 🎁
133
58
430
57,985
Thai Hoa retweeted
FAMILY, CHECK YOUR PORTFOLIOS 👀
Your welcome bonus is waiting. Check your balance and see when it unlocks.
A little something from us to get you started.
revenue.family/#portfolio
54,870
83,338
92,406
1,674,709
vậy là season 0 @arcus_xyz đã chính thức khép lại sau hơn vài tháng chạy thử nghiệm bản beta
hiện tại dự án đang live season 1 , có 2 cách tham gia để cày vol là :
+ trade spot stock
+ perpdex nhưng cần code hoặc phải nằm trong WL
Arcus là Perp DEX, kết hợp crypto, cổ phiếu, hàng hóa và RWA, hỗ trợ giao dịch long/short 24/7 trên-chain .
Arcus hiện được xây dựng trên Robinhood Chain, bởi đội ngũ đứng sau dYdX và hợp tác với Robinhood Crypto.
tham gia sớm tại đây để tìm cho mình vị thế sớm nhất nhé : app.arcus.xyz/ref/THAIHOA555
32
27
713