Bittensor Builders Are Years Ahead – Sami Kassab and David Lawee Explain Why on Oro’s Podcast

Oro (SN15) podcast pulled together one of the more substantive lineups in the ecosystem this year. Sami Kassab, founder of Unsupervised Capital and the analyst who coined the term DePIN at Messari, sat down with David Lawee, co-founder of Crucible Labs and the former head of a $7 billion fund at Google, alongside Oro co-founders Shardul Bansal and Seth Schilbe.

The conversation centered on what the agent capability gains of the last two months actually mean and why Bittensor is one of the few networks structurally built for the shift.

The teams who built quietly through the slower period before this are now sitting on a meaningful lead, and the argument that emerged is that this is not a coincidence.

Important Snippets

The conversation produced a tight set of takes worth pulling out individually:

a. Capability is no longer the bottleneck (Management is.) Shardul noted that he and Seth have shipped more code in the last two months building Oro than in nearly twenty years of combined software engineering before that.

The new constraint is keeping a swarm of agents pointed at the right work.

b. Agentic commerce needs evals before it needs scale. Closed-source labs have internal benchmarks for their agents, but commerce has almost nothing comparable.

Ridges‘ work on SWE-Bench-style evals for software engineering is the template for what the commerce eval layer will eventually need.

c. Digital commodities will lead the category. Software, compute, and content carry less downside risk per agent mistake than physical goods, which is where today’s brittleness bites hardest.

d. Cheap inference is the precondition for mass adoption. Open-source models through decentralized networks like Chutes are what make consumer-facing agentic commerce economically viable, since closed-source pricing cannot reach the price points users will actually pay.

e. Agent personalization structurally beats PageRank. Every buyer weighs price, timeline, availability, and quality differently, and sub-agents resolving those criteria deliver search precision traditional rankings cannot match.

f. A subnet is a financialized RL environment. Sami’s framing was that subnet owners define the objective function and miners compete to maximize it for tokenized rewards, which is the same structure frontier AI engineering already uses.

g. Bittensor uses only the valuable parts of crypto. Most crypto AI projects failed trying to attach on-chain transactions to every inference. Bittensor used incentive mechanisms and coordination of distributed labor instead, while fixing open source’s two structural problems of funding and contributor alignment.

h. The promise of crypto is small teams building significant things. David noted that Bitcoin was theoretically started by one person and OpenCloud was effectively a one-person billion-dollar company. Bittensor is the substrate designed to enable more of that pattern.

i. Opentensor Foundation was always meant to disband. The next iteration of stewardship comes from trusted actors like Crucible Labs, whose founding engineers still advise subnet teams like Oro on mechanism design.

j. Oro’s thesis sharpens the agentic commerce case. Roughly 100 factories in China produce the same goods with different logos for different brands, and an agent buying directly at the right point in the supply chain could deliver the same product at 10% of consumer price.

Where This Lands

The throughline across all four guests is that Bittensor’s structural design happens to match exactly what the agentic moment requires.

Cheap inference through decentralized model providers, financialized RL environments through subnets, founders treating the network as a creative playground rather than a token vehicle, and a stewardship layer designed to disband on schedule.

Oro is one of the cleanest tests of whether that combination produces real consumer products, with shopping agents as the proving ground. The brittleness of today’s agents and the missing eval infrastructure for commerce are real problems, but both are solvable, and the work is already happening across the network.

The teams who solve them first will define what agentic commerce looks like when it arrives at scale.