1/8/2026AI Engineering

Can Non-Technical People Code with AI Agents?

Can Non-Technical People Code with AI Agents?

The Rise of “Vibe Coding” and AI-Assisted Development

The concept of coding without being technical has gained significant attention with the advent of AI agents. Ben Tossel, a non-technical individual, has shared his experiences and insights on using AI agents to build various projects. According to Tossel, he has spent three billion tokens in four months, exclusively through a terminal, watching an agent write code that he couldn’t write himself. This approach has allowed him to ship numerous projects, including a personal site, a social tracker, and a crypto tracker.

Tossel’s method involves using a CLI (Command-Line Interface) agent called Droid, which is part of Factory CLI. He starts by discussing his ideas with the model, feeding it context about what he’s trying to achieve. Then, he switches to spec mode to plan out the project. Tossel emphasizes the importance of questioning assumptions and understanding the project’s requirements. He links relevant documentation and GitHub repositories to the agent, allowing it to explore and learn from them.

Key Takeaways from Tossel’s Experience

  • Tossel uses the CLI exclusively, as he finds it more capable than web interfaces.
  • He interacts with the model, feeding it context and questioning assumptions.
  • Tossel uses spec mode to plan out the project and identify potential issues.
  • He links relevant documentation and GitHub repositories to the agent.
  • Tossel watches the stream of code being written and intervenes when necessary.

Tossel’s approach to coding with AI agents involves building ahead of his capabilities and failing forward. He identifies gaps in his knowledge and thinks about how to improve his understanding of the system. This process allows him to learn and adapt as he works on various projects.

The Role of Agents.md in AI-Assisted Development

Tossel discusses the importance of agents.md, a simple open format for guiding coding agents. This file serves as an instruction manual for the agent, providing context and instructions on how to work on a project. Tossel has been experimenting with different agents.md setups to improve his workflow.

Some key aspects of agents.md include:

  • Explicitly setting up new repositories with instructions on what to do and not to do.
  • Specifying how to handle GitHub interactions, such as committing changes.
  • Defining whether to use a personal or work GitHub account.
  • Configuring testing and end-to-end testing procedures.

By using agents.md, Tossel can ensure that his AI agent follows a predictable and consistent workflow.

Coding on the Go with AI Agents

Tossel highlights the benefits of using AI agents to code on the go. He installs the Droid GitHub app on every repository he creates, allowing him to trigger the agent to review and make fixes from issues or pull requests. This capability enables Tossel to code from his phone and make changes when he’s away from his desk.

Tossel also uses Slack with his agent, creating a new channel for each repository and firing off tasks as needed. This approach allows him to manage multiple projects and ideas simultaneously.

New Skills and Knowledge Acquired

Through his experience with AI agents, Tossel has acquired new skills and knowledge, including:

  • Bash commands: Tossel learned to use bash commands to interact with the operating system and automate tasks.
  • CLIs: He prefers using CLIs over MCPs (Multi-Command Palettes) as they are simpler and more efficient.
  • VPS (Virtual Private Server): Tossel learned about VPS and how to use it to run his crypto tracker and sync his local repositories.

Tossel’s experience demonstrates that non-technical individuals can learn to code with AI agents and acquire new skills in the process.

The New Programmable Layer of Abstraction

Tossel discusses the concept of a new programmable layer of abstraction, where the focus shifts from mastering drag-and-drop tools to working with AI agents. This new layer requires understanding how to prompt the model effectively, provide the right context, and improve the system over time.

Tossel notes that this new paradigm feels like a game, where he can build and explore various projects without being limited by his technical knowledge. He emphasizes the importance of asking “silly questions” and being a student of the coding world.

Conclusion

Tossel’s experience shows that non-technical individuals can code with AI agents and build various projects. By leveraging AI agents and learning from their outputs, individuals can acquire new skills and knowledge. The key takeaways from Tossel’s experience include the importance of using CLI agents, creating agents.md files, and coding on the go.

For those interested in exploring AI-assisted development, Tossel recommends starting with a CLI agent like Droid and being open to learning and experimentation. As the field continues to evolve, it is likely that we will see a proliferation of software development, with both good and bad projects emerging.

To learn more about the latest developments in AI and its applications, you can refer to The Evolution of AI: A Deep Dive into the Latest Developments.