Summary of the full stream (approx. 1 hour) "The Infrastructure Behind Agentic Finance" with Shashank Yadav (founder of Fraction AI) and Greg Osuri (founder of Akash Network).
Main topic:
Can AI agents save/scale crypto and DeFi?
Both speakers' answer: Yes, they can — but only if the infrastructure is built for agents (crypto-native), not for humans.
Key ideas by section:
1. History of Akash Network and decentralized cloud
Greg Osuri started with open-source (Kubernetes, Terraform), and saw how AWS and other clouds turned open-source into a rent-seeking economy (50–70% of SaaS budgets go to cloud providers).
The idea of Akash: a decentralized supercloud — using unused datacenter capacity around the world.
Launched in 2020, initially focused on GPUs for AI (even before the ChatGPT boom).
Currently on Akash: ~10–15k GPUs (mostly enterprise-grade, not home 4090s), utilization ~70%. Demand is growing, but supply is limited.
2. Problems with decentralized compute
Reliability is the biggest pain point: nodes can drop out, especially home GPUs.
Energy: datacenters are hitting electricity shortages. They're trying a "Home Node" program (connecting home 4090/5090s).
Fault tolerance: for AI tasks, they use smart routing and task recovery when a node fails.
Future: long-term contracts, large clusters (50–100 nodes with H100s), distributed training (currently immature, but evolving).
3. Main part: AI agents and crypto/DeFi
This is the most relevant section (approx. 30–50 minute mark).
Crypto for humans has terrible UX (wallets, seed phrases, gas, bridges, KYC, etc.).
For AI agents, crypto is the ideal environment:
- No bot protection
- No logins and passwords
- Just a private key smart contracts
- An agent can find endpoints, perform DEX swaps, manage positions, find yield, etc.
Example from Greg: Akash Console already allows agents to deploy workloads for tokens without human involvement.
Centralized platforms (AWS, banks) are, on the contrary, very inconvenient for agents (bot protection, verifications, KYC).
Conclusion: agents will be the primary users of crypto in the future. Humans may stay at the "entry point", but from there an agent will do everything.
Key phrase from the stream:
"Crypto wasn't built for us (humans). It was built for agents."
4. Problems and realism
Branding: Akash tries not to mention "crypto" in marketing to enterprises — they sell "cheap private compute," not a blockchain. Many deals were lost precisely because of the requirement to pay with AKT.
The AI community is often skeptical of crypto (considers it a scam).
Solution: focus on real utility (price, privacy, distribution), not ideology.
5. Future (predictions)
Distributed training of models will become mature in 2–3 years.
Home GPUs (4090/5090) will become widely used thanks to fault-tolerant systems.
Agents crypto = true "agentic finance": autonomous portfolio management, yield farming, 24/7 trading without human intervention.
Possible revival of metaverses through human-agent interaction.
Final takeaway from the stream:
AI agents can save crypto from the problem of low real DeFi adoption because they:
- Are not afraid of complex UX
- Can operate completely autonomously
- Need exactly the trustless, permissionless rails that crypto provides
But for this, the right infrastructure (like
@akashnet for compute
@FractionAI_xyz for agent training) and a focus on agent-first design are needed.
Will AI agents save crypto?
There are a billion people using crypto, but barely anyone actually uses DeFi to earn yield on their capital.
That’s because crypto UX is still pretty complex for humans. AI agents, on the other hand, cruise through it.
Our founder
@0xshai and
@akashnet’s founder
@gregosuri dive into this and more.
Full convo linked in the comments.