For the past few years, the crypto community has been eagerly anticipating the arrival of Web3, where users own their data, everything operates on the blockchain, and access to information requires a crypto wallet.

The standard vision of the next generation of the internet is based on decentralization. All information is not stored on centralized servers and is not used for personalized advertising.

But what if Web3 turns out to be a bit different? Let’s explore the future through the lens of decentralization achieved by creating personal, locally functioning applications with the help of AI agents.

What Web3 Is Expected to Be Like

With the emergence of cryptocurrencies and the development of blockchain technology, many have begun to view them as the foundation for the next generation of the internet. For instance, Tim O'Reilly, the author of the Web 2.0 concept, believes that Web3 could become a significant developmental stage if it learns to connect the crypto economy with the real world—encompassing legal systems, property, payments, identification, application services, and production.

The crypto community sees the main difference between the new generation of the internet and Web 2.0 as a deeper decentralization at all levels, including data storage and application functionality. Ideally, product development would be managed not by an owner but by a distributed community overseeing the project through a DAO.

Decentralization is viewed as a fundamental principle that has enabled cryptocurrencies and smart contracts to carve out their place in the economy: they reduce reliance on intermediaries and centralized structures.

Be Your Own Programmer

The development of large language models (LLMs), AI agents, and vibe coding allows for a new perspective on Web3. ForkLog does not advocate abandoning the ideas of decentralization and blockchain; rather, it suggests expanding the vision of the internet of the future.

What if every user could be their own programmer? Such a user could write applications not for public use but for personal tasks, run them locally on their own computer or a remote server, and remain independent of centralized providers.

Take decentralized exchanges like PancakeSwap or Uniswap. They consist of a set of smart contracts operating on the Ethereum, BNB Chain, and several other blockchains. The networks themselves are decentralized, and access to services is facilitated through non-custodial wallets.

It seems like this is already Web3. However, there remains a point of failure—the frontend. The official websites through which users access exchanges are still centralized: they can restrict certain tokens or block users by IP and address.

While it is possible to access smart contracts directly outside of the official site, it is cumbersome. Users can utilize third-party frontends, which again leads back to centralized points, or open the contract in a blockchain explorer like Etherscan and invoke functions through Write Contract. This is inconvenient, complex, and requires technical skills. Not everyone can manage it.

However, thanks to AI, there is a third option—write an application using vibe coding and run it locally on your PC. We attempted to create such a product using Zed, OmniRoute, and LLMs from Anthropic and OpenAI.

The frontend for the project was created through Lovable. When launched locally, the application currently looks a bit rough and requires interface improvements, but it performs all functions.

The application was developed in just a few hours of vibe coding without any programming knowledge. In the future, AI will become smarter and be able to generate ready-made tools without the need to write dozens of prompts and constantly adjust results. Perhaps a simple request like, “Create and deploy an application for providing liquidity on Uniswap” will suffice.

The idea of launching local applications can be expanded as far as the imagination allows:

  • Trading bots for decentralized exchanges—an algorithm could be written to search for patterns or execute trades from a crypto wallet, with communication facilitated via a Telegram chat bot for convenience and tracking results;
  • Services for utilizing lending protocols—similar to working with DEX, but in this case, the interface would allow users to deposit funds in Aave, Compound, or Venus and withdraw them in just a couple of clicks;
  • Interfaces for accessing decentralized social networks or censorship-free messengers—while it’s harder to envision how this would look in practice, why not?

Mobile applications are also an area where artificial intelligence can make an impact. It could write not only a website but also an APK file for Android with direct connections to smart contracts on the blockchain.

Imagine this scenario: you find out that the Spark service offers 12% annual returns on DAI stablecoins. You visit the site, but you are blocked by IP. Turning on a VPN doesn’t help. In the described Web3 future, this is not a problem. You open Claude Code and write a prompt:

“Create an application for earning through the Spark protocol on the Ethereum network. It should allow adding DAI and withdrawing them, as well as a dashboard for tracking investment performance.”

The AI creates a service for direct connection to smart contracts, bypassing frontend blocks. It runs locally on your PC—no centralized solutions involved.

Local AI

In such a Web 3.0, the single point of failure could be the AI itself, specifically centralized language models. ChatGPT, Gemini, and similar solutions operate on servers owned by OpenAI, Google, and other labs. They can filter traffic, impose censorship, and apply restrictions.

However, there is a solution even in this case—open-source LLMs that can be run on your machine or a remote server.

For example, you could set up a configuration like this:

  • Ollama—runs LLM locally on Mac;
  • OmniRoute—a router/proxy between Zed and the models;
  • Zed—an editor that connects to OmniRoute.

As a result, communication in Zed occurs like in a regular chat bot, it writes code and launches applications, while the LLMs operate locally.

Which model to choose depends on the machine's specifications. For instance, on a MacBook Air with 16 GB RAM, suitable models include qwen2.5-coder:7b, qwen3:8b, llama3.2:3b, deepseek-r1:8b. A more powerful model can be installed on a local server, but that’s not a free option.

There are many free powerful open-source models available, but most are Chinese—DeepSeek, Qwen3.5 from Alibaba, Kimi K2/K2.5/K2.6. Among American options, only Meta has tried to move in this direction, but the latest LLM was released as closed-source. Google has a Gemma line, but it is not flagship. Nevertheless, the neural network is good for local deployment.

In May 2025, Tether announced a new platform for developing "infinite and ubiquitous intelligence," which envisions the "launch and evolution" of AI agents on user devices instead of the data centers of large companies.

The QuantumVerse Automatic Computer (QVAC) eliminates the need for cloud connectivity and provides greater privacy, autonomy, and resilience. Its modular architecture allows developers to create and expand applications using small composite elements.

A peer-to-peer network ensures direct communication between devices and collaborative work without reliance on centralized servers.

Apple is developing AI focused on local operation on devices—Apple Intelligence. Some tasks are performed directly on the iPhone, iPad, or Mac to account for the user's personal context without collecting personal data. However, for complex tasks, it still uses its cloud—Private Cloud Compute. Apple claims that only relevant data is sent there, and after processing, it is deleted, with the system built around privacy.

Open Projects

In addition to writing your own code from scratch, you can always leverage existing open-source projects. Fortunately, there is GitHub, where many different implemented ideas can be found.

Here are a few projects for managing liquidity:

When searching for interesting repositories, it’s important to analyze the code, check contract addresses, and test their functionality in a test environment or with small amounts. There are no guarantees of quality implementation.

Ready-made solutions can either be applied as is or adapted to your needs. Often, there is no need to write code manually—just instruct an AI agent to make the necessary changes to an existing project.

Drawbacks

The main challenge in creating the described Web3 future is the lack of ready-made user-friendly solutions and the technical complexity of implementation. Writing a frontend for decentralized Web3 projects can be done today with the help of AI, but it remains difficult for the average user. Without support or hours of analysis of various resources, one cannot navigate vibe coding, installing tools like Zed or Antigravity, or launching local LLMs and connecting them through OmniRoute.

There is an option to use ready-made applications from OpenAI (Codex) or Anthropic (Claude Code), but in that case, the idea of decentralization loses its meaning, and significant expenses will be incurred for tokens. In the first scenario, theoretically, coding could be done completely for free by connecting several Google accounts to services that provide free tokens.

This could be one of the possible directions for the development of Web3:

  • everyone writes ready-made applications for themselves with the help of AI, rather than relying on third-party centralized companies;
  • everything is stored locally on devices or remote servers;
  • the necessary infrastructure is provided by decentralized blockchains and smart contracts.

It is still difficult to say whether technological advancements will lead to such a model. It is possible that Web3 will turn out to be more familiar—without a single point of control and with less dependence on large platforms, but still with decentralization provided by limited groups of developers. Meanwhile, users will primarily work with ready-made solutions.