Key insights

• Bittensor distributes roughly $100 million per year in crypto tokens to incentivize developers building AI products on its network
• Subnet 62 (Ridges) claims to match or beat Claude Code on coding benchmarks at one-fifth to one-seventh the price
• The hosts have a direct financial stake in the projects they are promoting, which colors the entire conversation
• Decentralized AI faces real hurdles: unproven benchmarks, thin liquidity, regulatory uncertainty, and quality-vs-cost tradeoffs

---

In Brief

Jason Calacanis and Lon Harris welcome Mark Jeffrey of Stillcore Capital and Ala Shaabana, co-founder of Bittensor, to make the case for decentralized AI. Bittensor is an open-source blockchain protocol that pays developers in crypto tokens ($TAO) to build AI products. The episode showcases three subnet projects that claim to rival centralized competitors at a fraction of the cost: a coding assistant, a private inference service, and a decentralized storage platform. The conversation is bullish throughout, but there is an important caveat: Calacanis is a partner in Stillcore Capital, the fund investing in the very projects being discussed.

---

The central claim

The episode's core argument is that decentralized networks can build AI products that match or beat those from well-funded companies, at dramatically lower cost. Mark Jeffrey frames Bittensor as doing for AI talent what Bitcoin did for energy: creating a global, open marketplace where anyone can contribute and earn without needing approval (05:10).

The mechanism works like this: Bittensor operates a blockchain that distributes roughly $100 million per year in TAO tokens to "miners." But unlike Bitcoin miners who solve math problems with no practical output, Bittensor miners compete to build and improve actual AI products (06:13). These products live on "subnets," which function like specialized companies within the network. There are currently 128 subnets, each focused on a different AI task.

Ala Shaabana, who co-founded Bittensor in 2019 and now runs Crucible Labs (a wallet and allocation platform for the ecosystem), describes the network as evolving toward full decentralization. He and co-founder Jacob Steeves have both stepped away from the Open Tensor Foundation, handing governance to the community (35:39).

---

Three showcase projects

Ridges: a vibe coding competitor

The biggest claim in the episode comes from Subnet 62, called Ridges. Jeffrey describes it as a "vibe coding" platform (a tool that lets users describe what they want in plain language and the AI writes the code) that scores 73-88% on SWE-Bench (a widely used benchmark for evaluating coding AI) and 96.3% on the Polyglot test (07:26). He claims these results are "comparable or better" than Claude Code.

The pricing is striking: $29 per month, compared to roughly $150-200 for Claude Code or Codex subscriptions. Jeffrey argues that Ridges was built for roughly $10 million in token emissions from the chain, while Cursor had to raise against a $29 billion valuation to accomplish something similar (07:37).

The development model is competition-based. Miners compete to improve the product's code quality, with top miners reportedly earning around $50,000 per day in subnet tokens (09:27). Jeffrey frames this as a way for talented developers anywhere in the world to participate in the AI economy without needing access to Silicon Valley.

Targon: private inference

Subnet 4, called Targon, provides what the episode describes as "industrial-grade" inference (when an AI model generates a response to your input) with end-to-end encryption. The key selling point is privacy: your prompts are not used to retrain any company's model (20:18).

Targon uses trusted execution environments (secure hardware zones where data is encrypted even from the machine's owner) to guarantee privacy. The project has notable investors, including reportedly Tobi Lütke, the founder of Shopify, and Ram Shriram, one of Google's first investors (20:46).

Unlike Ridges, Targon operates as a traditional venture-backed company in addition to having a subnet token, creating a dual structure where investors can participate either through equity or through the token.

Hippius: decentralized storage

The third investment is Hippius, a decentralized storage service on Subnet 75. Jeffrey claims it is 400x to 4,000x cheaper than Filecoin (another decentralized storage network) (22:29).

What makes Hippius interesting economically, according to Jeffrey, is that the token's value is directly tied to how much the product gets used. You must lock up Hippius tokens to mine, creating steady demand for the token (22:49). This addresses one of crypto's classic problems: the disconnect between a project's utility and its token's price.

---

AI agents as crypto miners

One of the episode's more novel segments features Jeffrey demonstrating what he calls a "Vibe Miner," an OpenClaw agent he trained to mine cryptocurrency on Bittensor's Subnet 85 (VidAIO, an AI-powered video compression service) (27:33).

The agent uses Bittensor's own infrastructure to do its work: Hippius for storage and Targon for compute, keeping costs at roughly $10 per day while earning about $30 (30:16). Jeffrey admits the setup is profitable but not yet sustainable. After a 24-hour immunity period for new miners, more skilled competitors typically push him out of the rankings.

Calacanis sees broader implications. He argues that AI agents could automatically find and exploit price gaps across the Bittensor network, creating a relentless, 24/7 compression of costs across storage, compute, and inference (34:01). Shaabana tempers this slightly, noting that quality must balance cost: the cheapest option is not always the best.

---

Opposing perspectives

The conflict-of-interest problem

The most important context for this episode is financial. Calacanis is a partner at Stillcore Capital, the fund that has invested in all three showcase projects: Ridges, Targon, and Hippius (06:57). Jeffrey co-founded the fund. Shaabana is the co-founder of Bittensor itself and runs Crucible Labs, which profits from ecosystem growth.

Every participant has a direct financial incentive for Bittensor to succeed and for TAO's price to rise. While Calacanis does briefly disclose the Stillcore partnership, the episode functions more as an investment pitch than a critical analysis. There is no skeptical voice at the table.

Unverified benchmark claims

Jeffrey's claim that Ridges scores 73-88% on "the benchmark test" is vague. He does not specify which version of SWE-Bench, what task set, or whether results are independently verified. In the rapidly evolving coding-AI space, benchmark numbers without transparent methodology are difficult to evaluate. The comparison to Claude Code and Codex also conflates very different products with different feature sets.

Regulatory and liquidity risks

Subnet tokens currently trade on limited exchanges. TAO itself is on Coinbase, but the individual subnet tokens require specialized wallets like Crucible. As Shaabana acknowledges, each subnet token needs to pass the same KYC and legal scrutiny as TAO before major exchanges will list them (16:28). With 128 subnets, this process could take years.

The broader regulatory landscape is shifting. Calacanis references upcoming interviews with the SEC and CFTC heads about bringing crypto "back to America" (04:02), but this is still a goal, not law.

---

How to interpret these claims

Follow the incentives

The most reliable guide to interpreting this episode is to ask: who benefits? Every presenter stands to gain from increased investment in Bittensor. This does not make their claims false, but it means the viewer should apply a higher evidence bar than for a neutral analysis. The episode would be significantly more credible if it had included a skeptic or competitor.

The "cheaper than X" framing

Price comparisons like "$29 vs. $200 for Claude Code" or "400x cheaper than Filecoin" are attention-grabbing but potentially misleading. Lower price often reflects earlier stage, smaller scale, or subsidy from token emissions rather than real savings. Bittensor's $100 million per year in emissions is effectively a subsidy. The real question is what happens when that subsidy decreases through halving events (the first occurred in December 2025).

Decentralization as advantage vs. marketing

The episode presents decentralization as inherently superior. But for most users, what matters is whether a product works reliably, is priced fairly, and provides adequate support. Decentralization can provide censorship resistance and permissionless access, which are genuinely valuable properties. Whether those properties matter more than the reliability and polish of centralized alternatives depends entirely on the use case.

What strong evidence would look like

To move beyond investment pitch territory, Bittensor's subnet projects would need: independently verified benchmarks (not self-reported), sustained user growth outside the crypto community, transparent revenue and usage metrics, and evidence that products remain viable as token emissions decrease over time.

---

Practical implications

For developers

Bittensor offers a novel income model: contribute code improvements to subnet projects and earn crypto tokens instead of (or alongside) traditional employment. If the benchmarks hold, this could be meaningful for developers in regions with limited access to tech industry jobs. But token earnings can swing wildly in value, making this a risky bet.

For investors

The Stillcore Capital model treats subnet tokens like startup equity. The checklist Jeffrey uses to evaluate them (how big is the market, how strong is the team, what is the competitive edge) is familiar, but the risks are different. Subnet tokens have thinner liquidity, less regulatory protection, and more correlation to crypto market cycles than traditional venture investments.

For AI users

If Ridges or similar projects deliver on their promises, decentralized coding assistants could eventually offer a cheaper alternative to tools like Claude Code and Cursor. Private inference through services like Targon addresses a real concern about data privacy. But both products are early-stage, and "comparable to Claude Code" is a claim that requires ongoing verification as both products evolve.

---

Glossary

---

Sources and resources