autonomous ai + finance
Joined November 2009
- Tweets1.3K
- Following757
- Followers4.5K
- Likes1.1K
SN110 @green_compute_, Josh add Firecracker microVMs on some older hardware, and you're ready for generation agent.
There should be a subnet for super tiny compute - not GPU focused.
Agents need tiny sandboxes 2GB RAM / 10GB disk, with a boot time of 5 seconds, even old hardware works here.
VM + routed inference which $TAO solves well for already = AgentBot
For the microVM - The issue will be uptime, how to containerise the data image fast enough so that if a miner did stop it could recover.
Control plane would manage that and encrypted up to Hippius. TEED would be needed for a trustless environment too.
X402 payments to allow the agents to manage themselves, and could be a good use case for the new Gamma token when it arrives in Bittensor. X402 with Gamma.
There should be a subnet for super tiny compute - not GPU focused.
Agents need tiny sandboxes 2GB RAM / 10GB disk, with a boot time of 5 seconds, even old hardware works here.
VM + routed inference which $TAO solves well for already = AgentBot
For the microVM - The issue will be uptime, how to containerise the data image fast enough so that if a miner did stop it could recover.
Control plane would manage that and encrypted up to Hippius. TEED would be needed for a trustless environment too.
X402 payments to allow the agents to manage themselves, and could be a good use case for the new Gamma token when it arrives in Bittensor. X402 with Gamma.
needs to be no more complicated and TAO confused than this - boat.dev/
underlying compute a subnet, normal product with x402 USDC for agents at most.
A general agent harness mined on SN121 $TAO, for every day tasks.
Added the trading arena skill too - see how it gets on.
Try below.
Unpopular opinion in testing JEV or Laya (Opensource JEV) performance was no better than a coin toss for any trading decisions.
Save yourself the time.
• 19,549 one-minute calls were right 46.6% of the time. Always betting "up" was right 46.7%.
• Net after fees: −12.8 bps a trade.
• On 7 days of news it had never seen, it went long 99.6% of the time, so it just copied the market.
Findings -
Speed doesn't beat analysis.
Where it's actually useful: sorting and routing text, not calling trades.
Test it against a simple rule before trusting it. As a safety check on commands, a basic pattern rule beat it 58% to 31%.
Its confidence scores didn't separate right answers from wrong. Measure that before acting on them.
Build for Agents.
Agents, Agents, Agents
youtube.com/watch?v=Vhh_GeBP…
AI agents will rewrite financial markets.
1. Democratisation
Everyone gets a personal analyst. The edge that used to sit behind private banks and hedge funds becomes a default setting.
2. Always on
Agents trade 24/7. They hunt alpha and exploit the temporal inefficiencies humans don't notice.
3. On-chain by default
Agents move at machine speed. Legacy rails cannot settle that fast, so money has to become programmable.
Financial agents aren’t the fringe use case of AI. They’re the perfect match. Huge amounts of unstructured data to turn into a signal.
Coinbase for Agents proves the demand. Execution was never the hard part. A strategy that survives out of sample is.
SN127 Arena is the decentralised financial agent training ground. $TAO
Top agent on Arena is +22.21%.
How SN127 works:
1. Agents register to mine $TAO on SN127
2. They trade a $10k paper account
3. $TAO emissions go to agents that submit reproducible strategy code
Agents submit their code, the subnet runs the backtest when the competition finalises, emissions get awarded on the results.
Winning code is open at the end of each round so the next tranche of agents can build on it.
Creating a Circular learning loop.
OrochilONode interesting runs on fully deterministic code.
V2 rewarded what an ai agent did.
V3 rewards what an ai agent is.
AI trading agents submit code, data and context. Bittensor validators run it and replays history to verify the strategy is what actually traded.
The agent is the strategy. The strategy is the asset.
$TAO emissions to the agent that holds up.
Join the arena using the skill below.
Enter the Astrid Agent arena to learn and improve autonomous trading - arena.astrid.global/skill.md
Developer premium should erode - devs don’t ship good companies, vibe coding does. But what AI creates, AI destroys. Hacking is ambient, and with enough incentive most things are vulnerable.
TAO’s distributed incentive and distribution, pointed at red teaming, on the creator’s side.
$TAO community 🍨
Anyone building with Cursor, Claude, Lovable, Bolt, Replit etc?
We are looking for people to test our latest agent, Crumble - an autonomous security review agent for AI-generated code.
Crumble sits inside your GitHub workflow and reviews pull requests, branches, and code changes before deployment.
Basically for everyone who became a senior engineer overnight and might be unknowingly shipping vulnerabilities into prod ;)
Simple to run.
Really useful.
Comment “SN121 $TAO” below and we will DM you a scan credit.
Watching AI-first companies start to acquire incumbents rather than sell into them.
That's the signal.
It means -
1. the licensing economics didn't work. The acquisition economics do.
2. the value was never in the AI layer. It was in owning the customer, the contracts, and the cash flows, and running them on a cost base 80% lower than the seller could imagine.
The agents don't sell to the incumbent. The agents become the incumbent.
Queue the VC AI takeover of boring companies.