Your Simple Guide to Ridges (SN62)
Software engineering is one of the most expensive parts of building technology. Companies worldwide spend hundreds of billions of dollars on developer salaries each year. Yet much of the daily work involves repetitive tasks like fixing bugs, writing tests, and resolving code issues. Those tasks are essential, but they are also slow and costly when done manually. Ridges tackles this problem by building a decentralized network of autonomous AI coding agents on Bittensor, where open competition drives rapid improvement and anyone can contribute.
What is Ridges?
Ridges is a decentralized AI subnet running as Subnet 62 on Bittensor. The project was originally launched under the name Agentao before being rebranded to Ridges AI. It operates as a competitive marketplace for autonomous software engineering agents that solve real-world coding problems end-to-end.
The core thesis behind Ridges is that the role of a human software engineer can be broken down into smaller, discrete tasks. Instead of relying on one large AI model to handle everything, Ridges trains specialized agents to master individual steps such as fixing code regressions, writing unit tests, or resolving GitHub issues. These agents then work together to deliver complete coding solutions. As a result, the platform does not replace one tool with another. It replaces the workflow itself.
Ridges describes its mission as building AI software engineers that fully replace human coders, from start to finish. The project sits within the broader trend of autonomous AI agents. However, it focuses specifically on software development rather than general-purpose AI assistance. According to the founder, the global engineering services market is worth approximately $400 billion, making the addressable opportunity substantial.
The documentation describes Ridges as a platform where users can submit an entire software problem and reliably expect it to be completed by AI agents. Instead of an engineer going back and forth with a model, the system handles the full task autonomously. This positions Ridges not as a coding assistant, but as a coding replacement layer.
How Ridges works?
Ridges uses a winner-take-all tournament model to drive competition among AI agents. Miners develop their agents and submit them as Python files using the Ridges CLI. Validators then pull the submitted code and evaluate it against standardized software engineering benchmarks. The primary benchmarks used are SWE-Bench and Polyglot, which test agents on real coding issues pulled from active GitHub repositories.
The evaluation process follows a multi-stage pipeline. First, submitted agents run through Screener 1, which performs a preliminary assessment to filter out low-quality submissions. Agents that pass move to Screener 2 for further evaluation. If an agent clears both screeners, it advances to three randomly selected validators for comprehensive testing. Validators execute agent code in isolated Docker containers, ensuring secure and reproducible evaluation. Each validator independently scores the agent on a mix of SWE-Bench and Polyglot problems.
A key design choice is that agents operate inside constrained sandbox environments during evaluation. They do not have access to the internet, which prevents them from looking up solutions or relying on external resources. The only external access allowed is to the Inference Gateway, a secure gateway that provides controlled access to AI model inference and embedding capabilities while enforcing strict cost limits.
The agent with the highest overall score gets manually reviewed by the Ridges team. If the code is free of exploits and is not a copy of another agent, it receives approval and earns 100% of subnet emissions until another agent achieves a higher score. This winner-take-all structure creates strong incentives to continuously improve.
All miner-submitted code becomes open source after successful evaluation. This creates a compounding effect where each new submission can build on previous work. Consequently, innovation accelerates across the entire network. The architecture is also model-agnostic, meaning it does not depend on any single large language model. Instead, Ridges functions as what the team calls a “thick agent layer” that integrates various open-source models to create complete coding solutions. This approach keeps inference costs significantly lower than alternatives that rely on expensive proprietary models.
Who is behind it?
Shakeel Hussein founded Ridges. Based in Toronto, Canada, Shakeel previously worked as an engineer at Supabase and was part of the transition team at Twitter during its acquisition. He initially pursued dentistry at James Cook University before switching to technology. During the COVID-19 pandemic, he taught himself to program. He later studied Computer Science and Physics at the University of Toronto.
Shakeel bought out his original partners from the Agentao project and relaunched it as Ridges AI with a new team. The project’s public GitHub repository lists 13 contributors, with several core developers working under the Ridges organization. As of late 2025, the team consisted of approximately six engineers, with plans to expand as product development accelerated.
Ridges has attracted institutional backing from multiple funds focused on the Bittensor ecosystem. DSV Fund made an initial investment of $300,000 in August 2025, followed by an additional $672,000 allocation. This brought their total position to over $1 million. Stillcore Capital, a U.S.-based hedge fund dedicated to Bittensor, has also invested in the project. Through a combination of OTC deals and direct token sales, the team has secured approximately $4 million in total funding, according to a public disclosure by the project.
In January 2026, Ridges announced a strategic partnership with Latent Holdings, a prominent Bittensor development company that contributes to key ecosystem tools including the btcli command-line interface and the Bittensor Python SDK. Latent Holdings also operates the TAO.app blockchain explorer and Subnet 14 (TAOHash). The goal of this collaboration is to accelerate product delivery by combining resources with a larger organization that has deep experience in the Bittensor ecosystem.
Why it is valuable?
Ridges has demonstrated rapid technical progress that stands out even in the fast-moving AI coding space. Within four months of launch, the project achieved open-source state of the art on the full 500-question SWE-Bench Verified set with a score of 73.6%. The top agent has since surpassed that mark, with scores continuing to improve as new miners enter the competition. According to the team, every single one of their 50 evaluation problems has been solved by at least one agent.
What makes these results notable is how they were achieved. Rather than spending hundreds of millions on proprietary model training, Ridges used its incentive mechanism to drive a community of independent developers to continuously improve their agents. The entire SWE-Bench evaluation can be reproduced by anyone for as little as $1.26 for all 500 questions, using Bittensor’s own Chutes subnet for inference. In contrast, centralized competitors like Cursor and Devin have raised hundreds of millions in venture capital to build closed-source solutions.
The open-source model creates a compounding advantage. Because every successful agent eventually becomes public, each new competitor starts from a higher baseline. This means that even losing agents contribute valuable learning to the next iteration. Furthermore, the model-agnostic architecture allows Ridges to use cheaper open-source models while still achieving competitive results. The team has specifically highlighted integration with Bittensor subnets like Chutes and Targon as a way to reduce compute costs dramatically.
Additionally, Ridges opens a direct earning opportunity for developers worldwide. Anyone can submit an agent and compete for emissions. The daily prize pool has reached as high as $55,000 at peak levels, distributed on a winner-take-all basis. Rewards stream out every minute, with the leading agent continuing to collect until someone submits a better one. Performance determines earnings, not location or affiliation.
The future of Ridges
Ridges has been working on its first consumer-facing product. The initial plan called for a Cursor-style VS Code extension priced at approximately $12 per month, with an optional lower tier around $8 for users who agree to share usage data. The team also built a proof-of-concept tool called @taogod_terminal that demonstrated AI agents submitting pull requests directly to open-source repositories.
The partnership with Latent Holdings, announced in January 2026, represents the most significant strategic shift in the project’s history. The Ridges team described it as a move to dramatically accelerate product delivery. According to the announcement, the subnet’s incentive mechanism and alpha token structure remain unchanged. What changes is the speed and quality of execution, backed by a combined team with deep experience shipping products within the Bittensor ecosystem.
The longer-term vision involves shifting from benchmark-based rewards to real-world usage-based incentives. Instead of miners earning emissions solely by scoring well on SWE-Bench, the product itself will determine who earns rewards. Miners would be compensated based on whether real users accept their agents code suggestions, need fewer fixes, and remain engaged over time. This transition is intended to align the network’s incentives directly with the quality of the end-user experience.
Beyond its initial product, the team sees potential in enterprise adoption. The open-source, end-to-end stack offers compliance advantages for organizations with privacy policies that restrict sending proprietary code to third-party providers. Combined with secure compute options available through other Bittensor subnets, Ridges aims to offer a viable alternative to closed-source AI coding tools for both individual developers and large organizations.
Sources:
https://docs.ridges.ai
https://github.com/ridgesai/ridges
https://www.ridges.ai/explore
https://x.com/ridges_ai
