{"slug": "the-dogfood-advantage", "title": "The Dogfood Advantage", "summary": "Anthropic and OpenAI are accelerating faster than competitors because they extensively use their own AI products internally, creating a compounding feedback loop that improves both models and business operations. This \"dogfood advantage\" gives them early access to new models, unlimited token usage, and tight integration across all knowledge work, unlike companies where builders and users are separate.", "body_md": "# The Dogfood Advantage\n\nPeople often explain OpenAI and Anthropic’s velocity with culture, talent or capital. I think there is a more mundane compounding effect at play: dogfooding.\n\nDogfooding is usually framed as a QA discipline. Using your own products is important for making sure they work well. I have noticed that the value of dogfooding is actually on a spectrum, and recently an extreme version of it has appeared.\n\nAnthropic ships Claude Code with Claude Code. Their marketing team uses cowork to dominate news all day. This is why we’re seeing such an acceleration from the top labs. I call this **the dogfood advantage**.\n\n(source: [anthropic blog](https://www.anthropic.com/institute/recursive-self-improvement))\n\nI have been hearing the term dogfooding since my early days in tech. When I worked at Pinterest it was often thought that we had a disadvantage vs Reddit or Twitter since the people building the product (the developers) were not natural DAUs. Some would force themselves to use the product, but the feedback loop was not nearly as tight.\n\nContrast this with Twitter or Instagram or Whatsapp, where the majority of developers are daily active users. I think it’s one of the underrated reasons that some technology products are significantly better than others.\n\nLet’s push the idea further: some companies build products whose goal is not entertainment, but rather productivity. What if you heavily use your own product and this usage gives you some kind of competitive advantage in business? Some concrete examples: Figma is better at design because they use Figma so well; Stripe is good at billing; Notion runs on Notion (this one might be the closest to the labs since they have all of the data for knowledge work and now have agent harnesses also).\n\nOn the low end of the dogfood advantage scale, we have companies where the customer is completely separate from the builders, say hospital software like Epic or education software. Intuitively you can see how those products wouldn’t get nearly as good (and wouldn’t create improvement flywheels), because they lack feedback loops which compound.\n\nAI labs are in an extreme position on this spectrum. The labs are producing models and harnesses which are intelligence on tap, fueling all parts of the business and being used by their entire workforce. The whole idea behind [the great convergence](https://nicholascharriere.com/blog/the-great-convergence/) is that all knowledge work will be performed by agents using AI models in automated loops and workflows. Their core product does this.\n\nThis means that the dogfood advantage of these companies is on another level.\n\nCan you think of any company where people use their products *more* than at Anthropic or OpenAI? I can’t. They are using their products all day (and all night now!), dogfooding, getting value, getting feedback. Better model -> better harness -> more internal productivity -> better model.\n\nThis advantage has many facets, from getting early access (imagine having access to Fable or 5.6 for months before everyone else) to being token rich, since token cost is a rate-limiter for everyone except the lab.\n\nRecently, Jarred blogged about his journey of using Anthropic mythos to rewrite bun from Zig to Rust. Notably in this [article](https://bun.com/blog/bun-in-rust), he explains that he spent over $150k *and* was using a model that others didn’t have access to, leveraging both of these advantages:\n\nJarred’s blog mentioning both the insane token cost and the early access\n\nThis advantage compounds. I think this is one concrete reason the labs are accelerating relative to the rest of the economy. It’s not the only reason (selling intelligence APIs is about as high-PMF as products get), but it’s the most underrated one.\n\nBeing fluent in Salesforce doesn’t help you build chips. Being fluent in Claude helps every function in the business.\n\nWe talk often about moats, which I think are disappearing. I think we should be reasoning more about deltas, velocity, and compounding advantages. I think the dogfood advantage is one of these.\n\nIf you’re building something, ask yourself: do I have a dogfood advantage?", "url": "https://wpnews.pro/news/the-dogfood-advantage", "canonical_source": "https://nicholascharriere.com/blog/the-dogfood-advantage/", "published_at": "2026-07-11 18:10:25+00:00", "updated_at": "2026-07-11 18:35:18.772929+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-products", "ai-tools", "ai-startups", "ai-research"], "entities": ["OpenAI", "Anthropic", "Claude Code", "Pinterest", "Figma", "Stripe", "Notion", "Jarred"], "alternates": {"html": "https://wpnews.pro/news/the-dogfood-advantage", "markdown": "https://wpnews.pro/news/the-dogfood-advantage.md", "text": "https://wpnews.pro/news/the-dogfood-advantage.txt", "jsonld": "https://wpnews.pro/news/the-dogfood-advantage.jsonld"}}