Tidewave: José Valim's new direction for AI developer tooling

Episode Overview

• The episode begins with a discussion of the value of in-person events, prompted by the hosts' recent visit to the hardware company Oxide.
• The conversation transitions to the primary topic of AI coding agents, exploring the different ways developers and non-developers interact with them.
• José Valim introduces his project, Tidewave, an AI agent that runs locally within a developer's environment to enable real-time testing and verification.
• The discussion delves into the philosophy of building effective AI agents, arguing that providing them with fundamental code execution abilities is more powerful than creating numerous specialized tools.

Key Concepts

• Value of In-Person Events: The podcast opens by highlighting the benefits of in-person (IRL) interactions for building relationships and gathering stories, using their internal conference visit to Oxide as an example.
• Divergent AI Workflows: There are different approaches to using AI agents based on user expertise. Expert programmers like José Valim prefer to provide detailed context and fix errors themselves, while non-programmers tend to rely on iterative prompting to let the agent do the work.
• AI for Rapid Prototyping: AI agents can be used to autonomously build proof-of-concept applications over a short period, allowing developers to explore new ideas with minimal hands-on coding.
• Local-First AI Agents: The core concept of José Valim's project, Tidewave, is running the AI agent on the developer's local machine, not in the cloud.
• Shared Context and Self-Verification: By running locally, the agent gains access to the developer's full environment—including the browser, database, and REPL. This allows the agent to test its own code, verify features as a user would, and debug errors in real-time.
• Fundamental Capabilities Over Specialized Tools: The most effective AI agents may not need a wide array of specialized tools for tasks like database queries or API calls. Instead, they primarily need the fundamental ability to write and execute code within the application's backend and browser contexts.

Quotes

• At 2:08 - "I favor the IRL." - Adam Stacoviak expresses his preference for in-person events and interactions over remote ones.
• At 22:28 - "José is a better programmer than both of us, so he can just fix things." - Jerod Santo humorously points out that their reliance on AI differs from José's because José has the expertise to manually correct the code.
• At 25:43 - "My what I would do during the weekend is to come to the computer from time to time, see if the agent was working, and just have it build like a different proof of concept." - José Valim describes how he uses coding agents for hands-off, exploratory prototyping of new ideas.
• At 50:40 - "We are asking coding agents to develop without a proper browser, without a REPL. So Tidewave give all those things as well." - José Valim explains that his tool aims to bridge the gap between the tools human developers use and what is typically available to AI agents.
• At 53:14 - "Almost all of our APIs is 'write code'... you can execute code in the context of the web application, you can execute code in the context of the web page. That's it." - José Valim reveals his core philosophy that AI agents primarily need the fundamental ability to execute code in the relevant environments.

Takeaways

• AI coding agents are most effective when they can operate within the developer's complete local environment, allowing for real-time verification and a tighter feedback loop.
• The future of powerful AI development tools may lie in providing agents with core code execution capabilities rather than a fragmented set of specialized tools.
• AI can serve as a powerful assistant for rapid, hands-off prototyping, enabling developers to quickly explore and validate new product ideas.
• A user's technical skill level fundamentally changes how they interact with and leverage AI coding tools, from direct code correction to iterative prompting.