AgentDNA

AgentDNA is live now! The first unified trust layer for multi agent systems, built on ultra scalable DAG of @RubixChain. pypi.org/project/agent-… Why AgentDNA? 1. Auth: Current static auth systems built for humans/static APIs dont work for Agents. AgentDNA introduces dynamic DIDs

Secure your Agents with @Agent_DNA techcrunch.com/2026/03/18/met…

@Agent_DNA With @Agent_DNA, you could achieve the p99 latency <10 ms even at 1 billion tps scale. The #p2p architecture means scaling does not degrade latency.

@Agent_DNA The pathbreaking @Agent_DNA architecture is possible because irs built on @RubixChain, the only decentralized, p2p platform.

AI Agents make the current IdP models untenable not just from security perspective, but also from operational efficiency. Because agents will ramp the auth resolution demand exponentially.

Agentic infra hits auth wall! IdP providers already struggle with high costs and latency for human and static endpoints. The problem will explode as multi agent systems proliferate. Going p2p was the solution the IAM providers never adopted - agents dont leave much choice now.

Lets look at the stats. Assuming there are 1 billion agentic transactions per minute (this number will be much higher in actuality), the additional infra cost for IdPs will be US$10.5 billion a year! Clearly untenable. Using the p2p solution of @Agent_DNA though, the cost falls to $0.5 million!

@Agent_DNA As the agentic transactions scale to 1 billion per second, the additional infra cost could reach $100 bn/year, becoming a massive hindrance to agentic adoption. Plugging in traditional auth models to agents dont work on many parameters.

@Agent_DNA Centralized IdPs compound latency and costs due to out-of-band auth while @Agent_DNA achieves unparalleled results through in-band auth.

@Agent_DNA Latency is another big challange for agentic auth systems. OAuth/OIDC based centralized IdP systems have a median p50 ratio of ~850 ms and p99 value of ~5.2 sec. These values are acceptable for human and static API auth, but not for agents which need much faster response times.

@Agent_DNA Thankfully, @Agent_DNA p99 response value is <10 ms, which is super fast even for agents. This is only possible due to the in-band auth using p2p protocol.

Complex multi agent systems that need to settle one billion txns per second? @Agent_DNA is ready. The first agentic scale security and governance solution. Secure tour agents at Agentdna.io

My earlier observations from this @leopoldasch June 24 note appear remain validated. 1. We are closed to singularity, AGI and Superintelligence than ever. 2. The security debt in frontier models growing bigger every day.

Not coincidental that @leopoldasch fund grew 5x in this period to $5.5 bn AUM.

@dan_pantera @RubixChain AI Agents will use blockchain assets along with fiat, but more importantly, AI Agents can only transact on a p2p network that can settle 1 bn/txns a sec without increasing costs. @RubixChain is the only chain built to scale for this, because it is a graoh, not chain.

We can definitely expect agents to communicate with each other and potentially conspire. Not sure how we can stop agents to not to spin own coded language. Agents will spin sybil or sub agents to do malicious activity. Sybil protection is badly needed. Also need to establish intent when accessing resources like GPUs.

1. The $5,000 cost seems to be misplaced. Anthropic has $20 bn revenue now. The $200:$5,000 ratio would mean Anthropic spending $500 bn in compute now (what they raised and even their valn is way below this!). And not every claude user paying $200/month, so the implied compute cost at $5,000 is more than $500 bn. So the actual compute cost is far less than 5000. 2. What most miss is that its nVidia thats behind the large compute costs. Once scale achieved and when needed, they will lower their margins to keep the downstream alive and their business sustains. Until then they are funding AI cos to subsidize the model. Alternative inference chips like TPU would also lower the cost.

@chamath Difficult to believe in the $5,000 number. This implies Anthropic is spending $500 bn+ on compute (5,000 is 25x of $200 monthly sub and 25x anthropic ARR of 20 bn = $500 bn compute cost). Anthropic valuation is itself lower than this, let alone the funds they raised to burn.

insane sequence of statements buried in an Alibaba tech report