Summary of Key Points
Pete Koomen, a partner at YC (a renowned startup incubator), led his team in transforming YC from an organization established before the AI era into an “AI-native entity” over the course of a year. They started by addressing the pain points of the financial team and built a comprehensive infrastructure for Agent (intelligent agents) that includes a unified database, a tool registry with more than 350 tools, and a skill system, all of which enable AI to improve itself continuously. The key highlights of this transformation are as follows:
- Non-technical employees can use natural language to operate AI and solve problems.
- Organizational knowledge is transformed into AI skills.
- A culture of transparency and trust has been established to create “organizational superintelligence.”
- They have challenged the misconception of “AI products without a core foundation,” advocating for integrating AI into existing software in a way that allows users to take control of the tools.
Detailed Breakdown and Interpretation
1. The Need for Change: The Inefficient Cycle of the Financial Team
YC’s previous work process was inefficient: the financial team would explain complex processes, engineers would write the code, and then the financial team would use that code, creating a repetitive cycle with low efficiency. At that time, Agent tools like Claude Code became popular, and Pete realized that he could use them to complete many tasks quickly, while the company was still sticking to the old methods. He wondered if the financial team could directly use natural language (rather than code) to operate AI and solve problems on their own.
The first breakthrough was the development of an “SQL query tool” that allowed finance staff, who didn’t understand coding, to ask questions in everyday language (for example, “Which space companies were invested in the last four batches?”). The AI would then generate the necessary SQL queries to search the database. Previously, such requests would take hours to process; now, they get instant answers—this is what they call a “magical moment.”
In simple terms: Instead of asking an accountant to calculate the expenses for the month, you can directly ask the AI, and it will provide the information for you.
2. The Three Essential Foundations for Making AI Useful
For AI to be truly effective in a company, three core foundations are necessary:
- Unified Database: All of YC’s data (information about invested companies, founders, financial records, etc.) is stored in a Postgres database, eliminating the need for scattered data in various third-party tools. This allows AI to access all relevant information at once and answer any business questions.
- Tool Registry: All tasks that can be automated using AI (such as managing office time, bookkeeping, event planning, etc.) have been turned into tools and listed in a shared repository. The number of tools has grown from 20 to over 350, covering all important work at YC.
- Skill System: These tools are categorized into “skills” (for example, “writing a two-sentence company description”). AI can combine these skills to perform tasks more efficiently. For instance, the financial team can use the “bookkeeping skill,” while partners can use the “coaching founders skill.”
In simple terms: The unified database acts as the AI’s “brain memory,” the tool registry as its “toolbox,” and the skill system as its “specialized abilities.” Together, these components enable AI to function like a versatile assistant.
3. How AI Improves with Use
YC’s AI skills can improve on their own. For example, the “writing a two-sentence company description” skill started with prompts written by partners, but then partners collectively coached founders, providing feedback to the AI. The AI has since optimized this skill based on that feedback and now performs better than any individual partner.
This is how “organizational superintelligence” is created: knowledge that was previously scattered among individuals is integrated through AI, becoming the collective capability of the entire organization. New employees can learn from the experience of all previous partners without spending months adapting to the company’s processes.
In simple terms: It’s like teaching a child to write; initially, you show them how to do it, and then they learn by observing others, gradually improving their skills on their own. AI learns and improves through the collective experience of the organization.
4. “AI Products without a Core Foundation”
Pete wrote an article titled “AI Products without a Core Foundation,” criticizing many current AI products. For example, adding an AI email-writing feature to existing software doesn’t give users the ability to modify the prompts—it’s like adding an engine to a horse-drawn carriage but still keeping the traditional structure.
He believes that true AI-native software should be “Agents wrapped around tools” rather than “old software with added AI features.” At YC, users can create their own prompts and adjust skill parameters, rather than being confined to pre-designed frameworks.
In simple terms: The old approach is like adding an autopilot button to a traditional car; the new approach is to build a fully autonomous vehicle where users can control how it operates.
5. The Key to Organizational Superintelligence: Transparency and Trust
YC made a bold decision to make all conversations with AI agents available to full-time employees. Initially, there were concerns about privacy, but it turned out that transparency:
- Encouraged mutual learning by allowing everyone to see how others use AI to solve problems.
- Fostered trust, enabling AI to access more data and perform better.
- Helped new employees quickly adopt the experience of experienced employees, thereby raising the overall level of the organization’s capabilities.
Garry pointed out that this requires a culture of equality and trust, which is not common in most companies but is easier to achieve in startups. The investment (about $100,000 to $1 million per year in AI tokens) is relatively low, yet it allows the company to move ahead of its competitors by adopting advanced technologies (like those expected in 2028).
In simple terms: It’s like an open kitchen where everyone can see how others cook, and new employees can quickly learn from the experts, improving the overall quality of cooking.
Conclusion
YC’s AI transformation didn’t involve buying ready-made tools; instead, they built their own infrastructure to integrate AI into the organization. The key elements are a unified database, open tools, and a culture that encourages AI to learn from collective experience. For companies looking to implement AI, the focus should be on unifying data, making tools accessible, allowing AI to learn from collective knowledge, and fostering a culture of transparency and trust. AI is not meant to replace people but to empower everyone with the wisdom of the entire organization, making it stronger as a whole.