Ridges AI subnet 62: the coding-agent competition on Bittensor

Most software projects carry a list of small jobs nobody has got round to, such as a function that fails on empty input or a test that was never written. Ridgeline by Ridges.AI lets you hand one of those jobs over as a GitHub issue. An AI agent works on it and opens a pull request, which you review before anything is merged.

That agent comes out of a competition. The Ridges subnet is number 62 on Bittensor, often written SN62, and it launched in late 2024 under the name AgenTao. At ridges.ai, developers submit coding agents, Ridges tests them against hidden problems, and an agent earns the subnet's rewards by beating the current leader.

From a GitHub issue to a pull request

Say your Python project has that empty-input bug. You describe it in a GitHub issue, the way you would for a colleague, and pass it to Ridgeline. It connects through a GitHub App that can only reach the repositories you choose.

The agent then works on its own in a space Ridges manages. It runs your existing tests, makes the change, and runs the tests again. Anything that passed before has to pass afterwards. Then it opens a pull request, and nothing is merged until you approve it.

The agent suggests code. You still decide whether it is right. Ridges says miners never see which jobs their agent ran, but parts of your code do go to outside AI model providers while the agent works. If your repository holds code you cannot share with outside providers, that matters more than the price.

According to Ridges' docs, read on 30 October 2026, one pull request costs one credit and a rerun costs another. A $9.99 monthly subscription buys 10 credits. You can also earn credits by locking SN62 Alpha, the subnet's own token. Every 1,000 Alpha you lock earns one credit every three days, and those credits expire after two weeks. Locked Alpha still earns staking rewards.

Ridges' roadmap says Ridgeline launched in beta in the first quarter of 2026, and we found no customer, user, or revenue figures published for it. Compute subnets such as the one in our Targon spotlight rent out machines. Ridges sells a finished piece of work, so judge it by how often its pull requests are good enough to merge.

How the competition picks the agent

Each agent is a single Python file written by a Bittensor miner, the name for someone who supplies the work on a subnet. Ridges gives the agent a code repository and a written task, and the agent has to return a fix. It runs in a sealed box, with a time and spending limit for each problem.

Ridges then checks the fix against hidden tests that miners never see. No AI model grades the code. A separate check rejects agents that try to game the system, for example by hard-coding answers.

The diagram below shows the five stages an agent passes to take the lead: upload, two screening rounds where it must solve 45% and then 60% of the problems, three validators testing it on 50 problems each, and a final comparison with the current leader. Validators are independent nodes on the Bittensor network that check the miners' work. Ridges says these figures come from one recent competition and may vary.

Passing those checks earns nothing on its own. A new agent has to score at least 3% higher than the current leader. The other way in is to cost at least 6% less to run while scoring at least as well.

Each competition tests one skill. On 30 October 2026 Ridges' competitions page gave Test Generation and Database Query Engineering a 48% share of SN62's rewards each, and Linting 4%. Problem Solving was marked "Draining", with uploads closed and a 0% share, and the page gave no reason.

How rewards and burns work

The agent on top earns the most. When a new agent takes the top spot, it starts earning, and those rewards halve every two weeks. Earlier leaders keep a fading slice. So builders keep improving their agents to stay on top.

Ridges decides how SN62's rewards are split between competitions, and it can change the split while they run. Any share not given to an active competition is burned, which means it is destroyed and never paid to anyone. On 30 October 2026 the active shares added up to 100%, so nothing was being burned this way. Uploading an agent also has a fee, starting at $5, and that fee is burned too.

Who runs Ridges, and who owns the agents

Ridges' roadmap says that in the first quarter of 2026 "Latent acquired Ridges". The ridges.ai site and the Ridgeline terms name Latent Holdings as publisher and operator. Neither the site nor the docs names individual team members.

Kraken's listing post calls the agents open source, which needs a caveat. Ridges' miner FAQ says that submitting an agent hands all intellectual property rights in it to Ridges.AI, permanently. The current leader's code stays private while it holds the lead. Once an agent loses the lead its code is published, but for non-commercial use only.

The code that tests agents and runs validators is public in the Ridges GitHub repository. A change merged on 29 October 2026 added code for Ridges to run validators of its own, on fewer than half the checks for any task. It is not confirmed that these are running yet, and the docs still say Ridges does not host validators.

Risks that are specific to SN62

Price impact is how far your own order moves the price. The less TAO a subnet's pool holds, the further a given order moves it, so check the price impact on your quote before you swap. Slippage, from other trades that land before yours, comes on top.

SN62's share of Bittensor's rewards depends on its average Alpha price compared with other subnets, so a falling price can mean a smaller share.

There is also deregistration, which means a subnet is removed from Bittensor. When all 128 subnet slots are full, a new subnet replaces the one with the lowest average Alpha price, apart from subnets in their first six months or so. SN62 is well past that. If a subnet is removed, its TAO is shared across all Alpha holders, so what you get back may differ from what your position was worth before. Subnet numbers are reused, so check for the name Ridges as well as SN62 before you swap.

Much of the control sits with Ridges. It sets each competition's share, owns the submitted agents and, if its own validators go live, will do part of the checking too. Any of these can change how the subnet's rewards flow.

The usual Alpha risks apply as well. Staking TAO into a subnet swaps it for Alpha, whose value in TAO can rise or fall. Your staking rewards depend on the validator you choose and its take, and an estimated APY is not guaranteed.

How to buy Ridges SN62 Alpha on PotaTao

You buy SN62 Alpha by swapping TAO for it from a Bittensor wallet you control. It comes from the subnet's pool, so you need TAO first. To pay with dollars or another coin, buy TAO and move it to your wallet. Kraken announced SN62 trading on 21 July 2026.

On PotaTao it takes five steps, and the Swap guide covers the interface in more detail.

1. Open PotaTao and select Connect Wallet. Connect the wallet holding the TAO you want to use.
2. Leave TAO in the Sell field. Open the Buy token selector and enter Ridges in the search.
3. Choose Ridges, SN62. It appears in the Buy field.
4. Enter the amount of TAO to sell. Check the estimated SN62, price impact, any fees shown, and the selected validator. Open Settings to check max slippage and MEV Shield.
5. Select Review, check the amounts and validator, then select Swap and approve it in your wallet only if the request matches.

PotaTao never takes custody of your tokens. Quotes and rewards are estimates, and MEV Shield helps protect supported swaps without removing every risk. What MEV protection does explains the limits.

Open PotaTao.