Redefining the Company with AI
- The concept of building a self-improving company with AI is based on the idea of reimagining what a company is and how it acts, rather than just adding AI as a tool to make existing processes more productive 10s.
- Traditional companies are often organized like Roman legions, with a hierarchical structure where human beings are the conduit for information flowing up and down, but AI can break this model by allowing for more efficient and automated information flow 2m6s.
- The old way of thinking about AI is that it can make engineers 20% more productive by acting as a co-pilot, but this is a limited perspective, and instead, AI can be used to reimagine the entire company and its processes 4m30s.
- Extracting domain knowledge from a company and defining it as a set of skills or context is a key idea, as this knowledge is currently scattered across various sources such as emails, Slack messages, and Notion documents 6m15s.
- By making this domain knowledge legible, a company can move from a hierarchical organization to an intelligent, AI-powered organization with AI-native software, where AI is not just a tool, but an integral part of the company's structure 8m0s.
The Structure of a Self-Improving Company
- A self-improving company can be built using recursive AI loops, which consist of a sensor layer, a policy layer, a tool layer, a quality gate, and a learning mechanism, allowing the system to improve itself without human intervention 10m0s.
- An example of such a loop is an agent that can query a database and learn from its interactions with the real world, allowing it to improve its performance over time, even when humans are not actively involved 12m30s.
AI as a Productivity Enhancer
- The use of AI in companies can make them more effective, with AI acting as a sidekick to enhance productivity, such as querying databases to find relevant founders for meetings, and this can make a group partner 20 or 30% more effective 10s.
- The introduction of a monitoring agent that reviews every query made by employees and identifies areas for improvement can lead to significant self-improvement, allowing the AI to update skills files, databases, and indexes overnight to ensure successful queries the next day 42s.
- By identifying parts of a company that can work in a self-improving loop with AI, such as product analytics or customer service queries, companies can eliminate the need for human supervision and throw tokens at the problem to improve the company 2m6s.
Examples of AI-Driven Self-Improvement
- Examples of self-improving loops include having an agent analyze product analytics to identify areas of friction, researching best practices, and deploying AB tests, or using an agent to triage customer suggestions and deploy new code overnight 2m6s.
- To achieve this, companies should focus on burning tokens, not headcount, and measure token usage to identify areas for improvement, rather than relying on headcount or traditional metrics 10s.
Measuring and Scaling AI Impact
- The use of AI in this way can lead to significant increases in revenue per employee, with companies potentially getting to demo day with 5x more revenue per employee than they did 18 months ago 10s.
- Middle management may become obsolete as AI takes over coordination problems, and companies may need to rethink their organizational structure, with two key roles being individual contributors (ICs) and builders or operators 2m6s.
- Companies should experiment with AI to the maximum to figure out what is possible and identify which employees are "token maxing" to determine where to focus their time and resources 42s.
Organizational Structure in an AI-Driven Company
- Middle management may become obsolete as AI takes over coordination problems, and companies may need to rethink their organizational structure, with two key roles being individual contributors (ICs) and builders or operators 2m6s.
- Companies should experiment with AI to the maximum to figure out what is possible and identify which employees are "token maxing" to determine where to focus their time and resources 42s.
- To build a self-improving company, it is crucial to have directly responsible individuals, rather than committees or groups, and to make the entire organization legible to AI by recording everything, such as emails, Slack messages, and office hours, 10s.
Making the Organization Legible to AI
- To build a self-improving company, it is crucial to have directly responsible individuals, rather than committees or groups, and to make the entire organization legible to AI by recording everything, such as emails, Slack messages, and office hours, 10s.
- Making the organization legible to AI means that every conversation, decision, and action should be recorded, as if it is not recorded, it did not happen, and this information can be used to create a self-improving system, 2m6s.
- The recorded information needs to be diorized, or aggregated and synthesized, into important parts, to provide the AI with breadcrumbs, and this can be done by categorizing the information into certain areas, such as fundraising, hiring, and co-founder disputes, 4m30s.
Diorization and Knowledge Synthesis
- The recorded information needs to be diorized, or aggregated and synthesized, into important parts, to provide the AI with breadcrumbs, and this can be done by categorizing the information into certain areas, such as fundraising, hiring, and co-founder disputes, 4m30s.
- An example of diorization is regenerating the YC user manual, which was written 5 to 10 years ago, by using 2,000 hours of recorded office hours to create a new, 150-page user manual that is dramatically better than the existing one, 6m20s.
- The user manual can be updated every month, and it becomes a self-improving, living brain of advice for founders, which can be used as context for an AI agent, allowing users to ask a super intelligent AI and get the combined wisdom of 16 YC partners, 8m40s.
Self-Improving Artifacts and Software
- Every function in the company should generate artifacts that can self-improve, and if it does not, it should be thrown away, and every function can create on-demand software, such as dashboards, using tools like Codeex 55, 11m30s.
- Internal operations teams should use this layer of intelligence to create their own dashboards and workflows, which should be considered disposable, while the data should be stored preciously, 13m10s.
- The approach to company data involves keeping all information, such as emails, in a format like markdown, and treating the software as ephemeral, meaning it can be generated and regenerated as needed, with the valuable part being the comprehension and knowledge inside people's heads 10s.
The Role of Humans in an AI-Driven Company
- The business context and skills are considered the valuable part of a company, while the software on top of it is seen as ephemeral, and as models get smarter, the software can be thrown away and regenerated with the original set of instructions 42s.
- In this context, humans play a crucial role in a company, sitting around the edge of the company brain, which consists of data, emails, skills, and knowhow, and interfacing with the real world, making contact with reality in places where models cannot reach, such as novel situations, ethical considerations, and high-stakes moments 1m30s.
- Humans are necessary for tasks that require emotional intelligence, empathy, and complex decision-making, such as sales conversations, and will continue to be essential for these tasks in the next 20 years 2m6s.
Building and Adapting to the New Company Model
- The idea is that if you were building your company today, you would start it with this approach in mind, and for small companies, it is possible to build it right from the beginning, while larger companies may need to rebuild and adapt to this new way of thinking 3m40s.
