just the title scares me. what do you mean - you can be a million dollar company without PMF? WTF? all the chase for PMF seems futile if you have big bro’s backing.
The $200/month threshold is interesting not just as a price point but as a filtering mechanism. The people paying it are the ones encoding workflows, running agents, treating it as infrastructure rather than a tool. That cohort's usage patterns are probably dramatically different from the $20/month subscriber in ways that compound over time. Curious whether you've seen signs that the infrastructure-mindset users are also the ones whose usage grows rather than plateaus.
I use Claude code for complex project - digitizing entire corpus of a magazine, making documentary movies of artists, the malleability is staggering; I don’t use it for code,
The pricing shift reads less like software vendors finding PMF and more like both labs admitting they're becoming utilities. Seat-based SaaS gave customers a predictable line item; consumption billing for an uncapped-productivity tool gives you Uber maxing its annual budget by March. That isn't a failure story, it's what metering looks like before anyone builds the discipline to read the meter. The deeper tell: when the harness itself commoditizes (OpenClaw passed 250k GitHub stars running locally for free), the only defensible margin left is the metered inference underneath it. Both companies repricing to API rates in the same quarter is them planting a flag on the one layer that can't go to zero.
Do you think the PMF is mostly with people who already know how to judge the output? I keep wondering whether the sticky part is the model itself, or all the little habits people build around checking it.
there’s something refreshing about anthropic describing a release as “a modest but tangible improvement” instead of pretending every model update is the second coming
As a small operator using Max, the subscription math is a pleasant surprise. The enterprise pricing shift you're describing makes that clearer. I'll enjoy it while it lasts.
I think you hit the nail on the head with the April inflection point of the AI providers finally finding the enterprise is going to be their cash cow.
Now that enterprises are hooked and need enterprise level agreements and features for LLM coding tools, it’s going to be hard to pull away from for those that started out with individual subscriptions or team plans as they move over to API/usage prices.
Oh, and another note. OpenAI has started to move where Codex is not only going to be a coding tool, but it’s their answer to Claude Cowork. They haven’t come out and said it directly, but you can see it in Codex’s general settings and the way that they’re presenting it in workshops to companies to be a tool for non technical users.
Codex's settings already hinting at the non-coding pivot is the tell worth watching. Worth noting the post's frame: enterprises aren't just locked into pricing, they're paying API rates because two-thirds are building agents in-house without a clear starting point, so the tooling becomes the default scaffold. Whoever owns the non-technical entry point owns that gap.
The check on that is OpenClaw: 250,000 GitHub stars in under four months, runs locally, no metered token bill to inflate. When the agent itself is free, a vendor steering you toward pricier reasoning paths just hands the work to the local model that doesn't charge by the thought. The padded bill is only safe to send while the alternative stays inconvenient.
I'd love to read Ed Zitron's http://wheresyoured.at/ take on this https://simonw.substack.com/i/199683963/enterprise-customers-are-now-paying-api-prices
just the title scares me. what do you mean - you can be a million dollar company without PMF? WTF? all the chase for PMF seems futile if you have big bro’s backing.
The $200/month threshold is interesting not just as a price point but as a filtering mechanism. The people paying it are the ones encoding workflows, running agents, treating it as infrastructure rather than a tool. That cohort's usage patterns are probably dramatically different from the $20/month subscriber in ways that compound over time. Curious whether you've seen signs that the infrastructure-mindset users are also the ones whose usage grows rather than plateaus.
Sanity in AI discussion. Kudos!
I use Claude code for complex project - digitizing entire corpus of a magazine, making documentary movies of artists, the malleability is staggering; I don’t use it for code,
but for work. 100x the throughput
The pricing shift reads less like software vendors finding PMF and more like both labs admitting they're becoming utilities. Seat-based SaaS gave customers a predictable line item; consumption billing for an uncapped-productivity tool gives you Uber maxing its annual budget by March. That isn't a failure story, it's what metering looks like before anyone builds the discipline to read the meter. The deeper tell: when the harness itself commoditizes (OpenClaw passed 250k GitHub stars running locally for free), the only defensible margin left is the metered inference underneath it. Both companies repricing to API rates in the same quarter is them planting a flag on the one layer that can't go to zero.
Great summary on the state of agents.
Do you think the PMF is mostly with people who already know how to judge the output? I keep wondering whether the sticky part is the model itself, or all the little habits people build around checking it.
Does product-market fit here mean people are paying for better chat, or that they’re starting to trust these things with ongoing work?
Do you think the PMF is showing up in switching costs yet, or are people still mostly chasing whichever model feels best this week?
there’s something refreshing about anthropic describing a release as “a modest but tangible improvement” instead of pretending every model update is the second coming
As a small operator using Max, the subscription math is a pleasant surprise. The enterprise pricing shift you're describing makes that clearer. I'll enjoy it while it lasts.
I think you hit the nail on the head with the April inflection point of the AI providers finally finding the enterprise is going to be their cash cow.
Now that enterprises are hooked and need enterprise level agreements and features for LLM coding tools, it’s going to be hard to pull away from for those that started out with individual subscriptions or team plans as they move over to API/usage prices.
Oh, and another note. OpenAI has started to move where Codex is not only going to be a coding tool, but it’s their answer to Claude Cowork. They haven’t come out and said it directly, but you can see it in Codex’s general settings and the way that they’re presenting it in workshops to companies to be a tool for non technical users.
Codex's settings already hinting at the non-coding pivot is the tell worth watching. Worth noting the post's frame: enterprises aren't just locked into pricing, they're paying API rates because two-thirds are building agents in-house without a clear starting point, so the tooling becomes the default scaffold. Whoever owns the non-technical entry point owns that gap.
The check on that is OpenClaw: 250,000 GitHub stars in under four months, runs locally, no metered token bill to inflate. When the agent itself is free, a vendor steering you toward pricier reasoning paths just hands the work to the local model that doesn't charge by the thought. The padded bill is only safe to send while the alternative stays inconvenient.