Subnet 36 – Autoppia: The Rise of a Decentralized Web Operator

By: @CryptoZPunisher

In Bittensor’s expanding ecosystem, some subnets are not just training models, they’re building the foundations of a new internet. One of the most ambitious is Subnet 36, driven by Autoppia, a project at the crossroads of AI, automation, and the web. Its mission: to create a new layer of decentralized intelligence capable of acting on the web like a human, but with the rigor of code, the speed of the cloud, and the incentives of TAO.

Automata: The Digital Human

Autoppia’s flagship product is Automata, an AI agent capable of controlling a web browser and performing tasks on your behalf. Clicking, filling forms, booking, navigating complex interfaces, all become automatable.

Each task spins up an isolated cloud browser instance where the agent executes the mission live. This makes automation traceable, transparent, and verifiable.

IWA: The Living Test Ground for Web Agents

At the heart of Subnet 36 is the Infinite Web Arena (IWA), a dynamic benchmarking environment that:

• Continuously generates realistic synthetic web environments
• Creates complex tasks (bookings, form submissions, deep navigation)
• Evaluates agents’ ability to reason, adapt, and generalize in new contexts

Unlike static benchmarks that encourage memorization, IWA continuously refreshes challenges, stimulating adaptive intelligence.

Proof-of-Value: A New Incentive Model

Subnet 36 operates on a Proof-of-Value (PoV) model: miners submit agents, they are evaluated in IWA, and top performers are rewarded with TAO.

This aligns perfectly with Bittensor’s design:

• Permissionless: anyone can submit an agent and earn rewards
• Open Benchmark: IWA is open, transparent, and dynamic
• Real-World Utility: solves practical web automation problems
• Interoperability: agents can leverage other subnets (language, search, vector DBs, etc.)

Dynamic Zero – The Breakthrough

On October 21, Autoppia is launching Dynamic Zero, a major upgrade that redefines how web agents are trained. Its goal: to break the illusion of progress and force models to think instead of just memorizing.

Most agents today memorize patterns instead of understanding them. Dynamic Zero changes this by demanding agents that adapt, reason, and truly think.

Dynamic IWA: When the Web Becomes Unpredictable

Dynamic Zero introduces four layers of complexity into IWA:

• D1 – Layout & Structure: randomized HTML layouts, divs, and containers disrupt static navigation
• D2 – Real-Time Data: fresh data generated every session ensures no two runs are the same
• D3 – Text & Label Variation: aria-labels, tags, and visible text constantly change
• D4 – Interactive Pop-Ups: modals appear mid-task, forcing agents to handle unexpected events

The result: memorization becomes impossible. Agents must reason, understand, and adapt, just like human operators.

Winner Takes It All: Incentivizing Excellence

Dynamic Zero also introduces a new incentive model: Winner Takes It All (WTA). Only the top miner in each evaluation cycle receives the full reward.

• Time Factor removed: execution speed no longer affects the score
• Efficiency Rewards removed: precision and task completion are now the sole focus

This model intensifies competition and drives every miner to aim for the top, creating a powerful engine for innovation and performance.

Infinite Web Arena Becomes a Full Platform

With Dynamic Zero, IWA evolves from a simple benchmark into a full ecosystem hub, structured in three sections:

1. Live Leaderboard (Bittensor): Real-time miner and validator rankings Historical performance data and SOTA comparisons (OpenAI CUA, Anthropic, Browser Use) Transparent evaluations where decentralized AI competes live against centralized systems
2. Webs & Use Cases Explorer (No Bittensor Required): Browse dozens of demo websites and use cases See what tasks agents are solving and how complexity evolves daily
3. Open Benchmark Access (No Bittensor Required): Test your own agents directly in IWA Choose websites and use cases, link your endpoint, and receive performance metrics

This democratizes experimentation. Independent developers, AI labs, and startups can train, test, and iterate freely even before joining the subnet.

What’s Coming Next

Dynamic Zero is just the beginning. Autoppia is already preparing the next wave of upgrades:

• More Complex Use Cases: harder challenges per website, closer to real-world logic
• Composite Tasks: multi-step missions requiring reasoning across multiple objectives
• Top-K Rewards: expanding rewards to encourage more agent diversity
• Open-Source Integration: allowing anyone to participate through public miner repos

Open Source, With Safeguards

Autoppia is transitioning toward a fully open-source Subnet 36, where anyone can build, deploy, and benchmark agents. Openness, however, comes with safeguards:

• Cloning Protection: if two agents achieve similar scores, priority goes to the current top miner. New miners must exceed the top miner by a clear margin (e.g., ≥5%)
• Multiple-UID Farming Prevention: identical agents deployed under multiple UIDs are treated as one and rewarded only once

These measures ensure an open, fair, and fiercely competitive environment, accelerating progress toward SOTA agents.

Why Autoppia Matters

Autoppia is not just building tools, it’s constructing a modular infrastructure for autonomous companies capable of running with minimal human input. By connecting Subnet 36 to other parts of the Bittensor ecosystem, such as Gradient, Lium, Tangram, or Chutes, it enables a future where decentralized AI and web automation converge.

In a world dominated by web interfaces, mastering their automation is a powerful advantage. Subnet 36 aims to be that advantage, a platform where agents don’t just execute tasks but learn, adapt, and collaborate.

Useful Links

• Website: autoppia.com
• Infinite Web Arena: Infinite Web Arena
• Autoppia Studio: Autoppia Studio
• Docs: Autoppia Docs
• GitHub: github.com/autoppia/autoppia_web_agents_subnet
• Autoppia Made Simple: What It Does, How It Makes Money, and Why It Beats the Big Players
• autoppia.substack.com/p/dynamic-zero-the-overfitting-punisher
• X: @AutoppiaAI
• Discord Autoppia
• Telegram: t.me/Autoppia
• Taostats: taostats.io/subnets/36/chart

Conclusion

Subnet 36 is more than an automation project, it’s an experiment in what it means to think on the web. With Dynamic Zero, the IWA becomes a living laboratory where agents compete, learn, and evolve against increasingly complex environments. This is not a race for size or speed, it’s a race for adaptive intelligence.

In a space where decentralized AI now directly competes with centralized giants, one thing is clear: Subnet 36 isn’t copying the old model, it’s reinventing how intelligence acts on the web.

Enjoyed it? A like and a share are the best ways to support my work. Thank you!