Key Takeaways from the Latest This Week in Startups Discussion

In a recent episode of This Week in Startups, host Jason Calacanis sat down with Mark Jeffrey and Bittensor co-founder Ala Shaabana to discuss the growing intersection of decentralized infrastructure, AI agents, and crypto networks.

The conversation focused heavily on Bittensor and its token TAO, exploring how the network is creating a new economic model for building AI products through programmable mining.

Below is a structured breakdown of the key points discussed in the podcast.

1. The Core Idea Behind Bittensor

Bittensor is a decentralized AI network that adapts the incentive model pioneered by Bitcoin but applies it to artificial intelligence.

Instead of miners solving meaningless cryptographic puzzles, Bittensor miners compete to produce useful AI outputs.

The system works like this:

• The network emits TAO tokens as rewards.
• Developers create subnets (specialized AI networks).
• Miners compete to provide the best AI performance within each subnet.
• The best contributors earn token rewards.

According to Mark Jeffrey, this effectively turns Bittensor into:

“Programmable Bitcoin mining for AI.”

The network currently distributes around $100 million annually in incentives to support the development of AI products across its ecosystem.

2. Subnets: The Building Blocks of the Ecosystem

At the center of Bittensor are subnets.

A subnet is essentially a specialized AI marketplace focused on a specific capability. Each one functions like a startup inside the broader Bittensor network.

Currently:

• There are around 120 subnets
• The system caps them at 128 to maintain quality
• Launching one can cost $100K to $1M depending on demand

Developers can build subnets for things like:

• AI coding tools
• machine learning inference
• video compression
• data storage
• AI marketplaces

Anyone can create a subnet, but it must compete for performance and adoption.

3. Example Subnet: AI Coding Platform Competing With Claude

One of the most notable developments discussed was Ridges (Subnet 62), a coding platform designed to compete with tools like Claude.

Key highlights:

• Benchmarks scoring 73–88% performance
• 96.3% on Polyglot coding tests
• Pricing around $29 per month
• 5–7× cheaper than comparable AI coding services

What makes this remarkable is how it was funded.

Instead of raising billions in venture capital, the project was funded through about $10 million in Bittensor token emissions.

This demonstrates a new funding model where protocol incentives replace traditional VC funding.

4. Another Subnet: Decentralized AI Inference

Another highlighted subnet was Targon (Subnet 4).

This project focuses on AI inference infrastructure, allowing developers to run AI models privately and securely.

Key features include:

• Private AI inference
• End-to-end encrypted computation
• Trusted execution environments
• Lower cost than traditional cloud providers

Unlike mainstream AI platforms, prompts are not used to retrain corporate models, making it appealing for privacy-sensitive applications.

5. Decentralized Storage: The Hippius Network

The discussion also covered Hippius, a decentralized storage subnet similar to Filecoin.

Hippius differentiates itself through:

• dramatically lower storage costs
• tokenomics tied directly to network usage
• staking requirements for miners

The token economics ensure that increased platform usage automatically drives token demand, addressing a common problem in crypto projects where the token has little real economic value.

6. Mining in Bittensor: Anyone Can Participate

Just like early Bitcoin mining, anyone can become a miner.

Miners contribute:

• compute
• storage
• model improvements
• infrastructure

Participants range from individual developers to large infrastructure providers.

For example, unused data center capacity could be deployed into Bittensor networks to generate additional revenue.

This model turns idle infrastructure into productive assets.

7. AI Agents Mining Crypto

One of the most fascinating experiments discussed was Mark Jeffrey’s use of AI agents to mine Bittensor networks.

Using autonomous AI tools (OpenClaw), his system:

• selects profitable subnets
• deploys infrastructure
• processes workloads like video compression
• earns tokens automatically

The system currently earns around $30 per day in rewards while costing roughly $10 per day to operate.

However, competition is intense — better miners frequently push weaker ones out of the network.

Jason Calacanis described this environment as:

“The most ferocious form of capitalism.”

8. Why This Matters for the Future of AI

Bittensor could fundamentally change how AI infrastructure is built.

Instead of relying on centralized companies like:

• OpenAI
• Anthropic
• Google

AI services could be built by distributed networks of developers competing globally.

This creates:

• lower costs
• global participation
• decentralized innovation
• open competition between AI systems

The model also enables talented developers anywhere in the world to earn directly from their work.

9. Infrastructure for the Ecosystem

The episode also introduced tools supporting the ecosystem, including a wallet built by Crucible Labs that simplifies participation in Bittensor subnets.

Such infrastructure is necessary because interacting directly with decentralized AI networks can still be technically complex.

Improving accessibility will likely be key to mainstream adoption.

Final Thoughts

The podcast highlighted a major shift happening at the intersection of crypto, AI, and decentralized infrastructure.

Bittensor’s approach introduces a new economic model where:

• AI services compete in open markets
• infrastructure is distributed globally
• developers are directly rewarded for improving models

While still early, the system could represent a new way to build AI — one where the network itself funds innovation.

As Jason Calacanis put it during the episode:

“This could become a relentless machine for lowering the cost of compute.”

If that vision plays out, decentralized AI networks like Bittensor could reshape the economics of artificial intelligence over the next decade.

WATCH THE FULL EPISODE BELOW: