Summary of Key Points
The most significant takeaway from this coding demo night in Silicon Valley is that the use of AI for writing code has evolved beyond the question of “whether it’s possible” to “how to utilize it.” People are no longer debating whether AI can write code (after all, someone used AI to create a 15,000-line LinkedIn client with tests in just one weekend); instead, they are focusing on how collaboration methods, tools, products, and even programming languages should be restructured when AI becomes a “full-fledged member” of a team. The widespread acceptance of AI-written code as the norm among the over 400 attendees at the event represents a crucial shift in perception—AI is no longer just an auxiliary tool; it has become a partner that works alongside humans.
Detailed Breakdown and Interpretation
1. Team Collaboration: Moving from “Reviewing Code After It’s Written” to “Aligning Decisions Before They’re Made”
In the past, the key to team collaboration was code review—people would check each line of code after AI had completed it. However, with AI writing code so quickly (hundreds of lines can be generated in minutes), manual review is simply not feasible. During the demo, Matt from ref.tools noticed that many people used AI to create plans but rarely shared them with their teammates. The problem lies in the timing of collaboration: it’s too late by the time the code is reviewed. The solution is to bring collaboration forward, aligning decisions before the code is even written. For example, ref.tools created a shared document where the team can use AI to draft plans, comment on them, and get approvals before having AI modify the code. HumanLayer took this idea a step further by developing a real-time diff viewer that allows teammates to see changes in real time and provide feedback directly, enabling them to adjust the direction of the code as it’s being written.
2. Tool Upgrades: Making AI’s Work Process Visible and Communicable
One of the main challenges with AI-written code is that it’s like a “black box”—you don’t know how it works or what changes have been made. The tools presented at the event aim to address this:
- Hunk: Allows users to view code changes in the terminal and directly comment on them (e.g., highlighting potential issues), effectively acting as a preliminary review by AI.
- SideShow: Turns code changes into easy-to-understand web pages, making even complex changes understandable, such as when an intern submits a PR for 100 files.
- Entire: Links AI session logs to Git, creating a “diary” of the AI’s work. This way, when similar issues arise in the future, both AI and new users can refer to previous conversations and decision-making processes.
The core idea behind these tools is to make AI’s work transparent, allowing humans and AI to communicate effectively through the code changes.
3. Product Evolution: From Human-Centric to Agent-Native
Traditional product interfaces are designed for humans, but AI may not understand them (for example, interpreting web page DOM structures or understanding screenshots). The goal now is to create “agent-native” products that can be used directly by AI without any additional learning:
- Alex modified the open-source repository OpenStatus by adding a chat interface, allowing AI to operate directly within it (check status, write applications) without the need for DOM parsing.
- Charming’s tip calculator is user-friendly for both humans and AI; anyone can use it to calculate tips, and AI can provide immediate updates (e.g., “How much total tip is due this month?”).
- NoInfra.ai goes a step further by making its AI agent accessible to non-technical users via Telegram. Mothers can use the agent to recharge their cryptocurrency wallets without understanding APIs or tokens.
The aim of these products is to make AI a “user” of the product, enabling ordinary people to utilize it with the help of AI.
4. Programming Language Innovation: A New Language Designed for AI
Current programming languages (like TypeScript) are designed for humans and focus on development efficiency. However, if AI is primarily responsible for writing code, efficiency becomes less important; accuracy takes precedence. One new language was demonstrated at the event:
- Uses “graphs” instead of plain text, making structural changes immediately apparent and more intuitive than manual diff comparisons.
- Automatically adds tracking data to the code, generating flame graphs that help AI identify performance issues (e.g., which functions take longer than 5 seconds to execute).
- Functions can be called directly within Python/TypeScript without the need for additional boilerplate code.
The logic behind this language design is that since AI will be writing the code, the language should be tailored to its needs: clear structure, easy checking, and automatic monitoring capabilities.
5. Democratization: Making AI Assistants Accessible to Everyone
Some of the most user-friendly demos focused on making AI tools accessible to ordinary people:
- The founder of Charming mentioned that his parents don’t use AI coding tools like Cursor but do use ChatGPT, so he created an app that can be integrated with ChatGPT for tasks like checking fridge inventory.
- NoInfra.ai’s founder developed an AI agent that allows mothers to manage their finances via Telegram—the agent can send emails, make phone calls, and even recharge their cryptocurrency wallets.
The essence of these demos is that AI is no longer a luxury reserved for programmers; ordinary people should also be able to have their own AI assistants to help them with tasks.
Conclusion
This demo night represents a “cognitive revolution”—people have accepted that AI writing code is the new norm, and the focus now shifts to “how to work together with AI.” Every aspect of technology, from collaboration to tools, products, and languages, is being redesigned to accommodate AI. This acceptance of AI as a integral part of our work processes marks the beginning of future technological changes. After all, when AI becomes a regular member of a team, our ways of working must also evolve accordingly.