@TechFlowPost: Y Combinator partner shares how to build an AI company? Current companies are like Roman legions; AI breaks this paradigm, restructuring the company into a recursive self-improving AI loop. Tom Blomfield's core advice: · Make the entire organization fully readable to AI (record everything...
Summary
Y Combinator partner Tom Blomfield shares how to build an AI company, proposing to restructure the company into a recursive self-improving AI loop, and emphasizing making the organization fully readable to AI and separating speaker information.
View Cached Full Text
Cached at: 05/26/26, 09:07 AM
Y Combinator Partner Shares: How to Build an AI Company?
Companies today are like Roman legions. AI breaks this paradigm, restructuring companies into recursively self-improving AI loops. The core practices Tom Blomfield suggests are:
· Make the entire organization fully readable to AI (record everything) · Perform speaker diarization, distilling massive amounts of information into breadcrumb clues usable by AI · https://t.co/kgSXOgIe3U
Similar Articles
@ycombinator: In a recent batch talk, YC General Partner @t_blom broke down how to build a self-improving, AI-native company. He walk…
YC General Partner @t_blom presented a talk on building self-improving, AI-native companies, emphasizing recursive AI loops and reducing headcount through AI automation.
@FinanceYF5: Tom Blomfield on How to Build a Self-Improving Company with AI
In his talk, Tom Blomfield discusses how to leverage AI to build self-improving companies, sharing relevant insights and experiences.
@manateelazycat: YC's video is worth watching. He hit the nail on the head sharing how AI-native companies should think about organizational structure. I really agree with several key points: 1. The company is a Roman legion, where individual soldiers are strong and there are no weak links, so they never fall into chaos during any tactical execution. 2. The co-pilot model is wrong; we should let AI do it completely because AI executes 10 to 100 times faster than you. If it makes mistakes, just redo it; the second and third attempts are still much faster than you writing code manually. 3. Burn tokens, not heads. That is so true. I think any AI-native company should give employees unlimited tokens. Like our company: many ask how we reimburse AI costs. We don't reimburse; we give everyone unlimited GPT 5.5 tokens, use as much as you want, no reimbursement process. Only with unlimited tokens will people use them and unleash their creativity without psychological burden. 4. Make everything clear and readable for AI. Let me interpret: previously you wrote documents manually, and people didn't want to. Now open up all document interfaces and MCP to AI as much as possible. When you open up to AI, it can use its organizational capabilities to create new company processes. Previously, building internal infrastructure systems was difficult due to huge workloads and many aspects. Now, as long as you expose enough MCPs and APIs, new processes are a matter of one sentence. 5. People are temporary, but contextual documents are important. That's also well said. In the future, the most important things for software companies will be requirement documents, context documents, and process documents. Because code is easy to recreate, but these documents and thought processes are the company's real assets. Alright, that's today's share. Welcome to like, bookmark, and repost. Follow me, I'm a veteran who has been working on operating systems for over 20 years. My daily job is browsing Twitter and reposting valuable content to save your time.
User shares agreement with the views from YC's video on AI-native company organizational structure, including burning tokens instead of heads, letting AI fully autonomously execute tasks, and the importance of documents as core assets.
@t_blom: A lot of people are trying to figure out recursively self-improving companies right now. I break down what we’re seeing…
A breakdown of recursively self-improving companies observed at Y Combinator, exploring how AI-driven automation enables companies to continuously optimize themselves.
@yibie: Recommended article: Eugene Yan from Anthropic (former Amazon/Alibaba ML team lead) writes a practical guide to his personal AI workflow. Not abstract ideas, but concrete methods you can replicate tomorrow: how to organize directories for easier model retrieval, how to write C…
This is a tweet recommending Eugene Yan's AI workflow guide, detailing how to efficiently collaborate with AI and achieve compound work gains by organizing context, configuring CLAUDE.md, creating skills, and more.