Daily Grind July 9, 2025: Cursor pricing fumble, Building Rocketships, and AI productivity?
Tech's morning newsletter, featuring one headline, one page of a great book, and one question to ponder
Hello!
Welcome to the Hump Day edition of The Daily Grind.
First, let me re-introduce myself since it’s been a while.
My name is Ben Putano and I’m the founder of Damn Gravity Media, a book publisher hyper-focused on business and tech titles. We love publishing books for entrepreneurs, by entrepreneurs.
We currently have 7 titles, with our 8th coming out on July 22!
I started The Daily Grind to bring together founders and entrepreneurs around the 3 things I love: startups, books, and thoughtful questions.
My goal is to not just give you the news, but use headlines to dive deeper into an important topic.
Case in point: Today’s headline is about Cursor’s pricing mishaps, but it’s really a conversation about the challenges of pricing AI products.
So I hope you enjoy, and if you want to read more Books for Builders, check out our store.
📰 One Startup Headline: Cursor fumbles pricing change (AI still hard to price)
Pricing AI products is still in its primordial soup era, and Cursor just cooked up something that made everyone wretch.
Cursor, the leading AI code editor, quietly changed its pricing on June 16. Some users realized this later when receiving massive usage-based bills they didn’t expect.
Cursor’s original pricing was simple and straightforward: $20/month for unlimited “Tab” autocomplete and 500 requests (a request is a prompt, like you would input to ChatGPT).
The problem was that this pricing model was costing Cursor money.
The challenge with AI pricing is that the costs of running LLMs don’t scale. AI companies must pay for every token (i.e., a chunk of output from an LLM) that users ask to generate. Simple requests require a few tokens, while complex request require a LOT.
In other words, Cursor’s pricing was fixed, but their costs were variable.
AI content writer, Finn Lobsien, covered the pricing fumble and Cursor’s conundrum on Substack:
“Cursor was stuck: They could either make pricing more complicated by distinguishing between simple and complex requests (and confuse users) or price directly to compute to protect their margins—which is what they did.”
New Pricing Plans
To protect its margins from variable costs, Cursor changed its pricing unit from requests to compute (i.e., tokens used). Their updated Pro plan ($20/month) included unlimited usage of their “Tab” model—an in-house LLM designed for autocomplete—and “rate limits” on frontier models usage (models from OpenAI, etc).
Cursor also introduced a new tier: the $200/month Ultra subscription, with 20x the usage rates on frontier models.
If users went over their rate limits (by making a lot of complex requests that required a lot of tokens), they were charged an overage fee. Plus, if users chose to use a more powerful model, (say GPT-4.1 instead of o4-mini), they would use more compute per request.
Like Uber Surge pricing, these fees surprised and angered users.
On July 4, Cursor shared a new post to clarify its pricing and plan changes. They clarified the term “rate limit” to mean “usage,” and offered refunds to any users who were surprised by their surcharges.
But Cursor may not be out of hot water. Looking at their current Pricing Page, the new compute-based pricing for Pro is still not clear.
Users have to click on “More Info” to get an explicit answer of what’s included in the Pro plan:
Unlimited tab completions
Extended usage limits on all models
Access to BugBot
Access to Background Agents
Cursor needs to get clear on their billing before more complaints go public. With OpenAI buying competitor Windsurf—and Claude launching Claude Code—they are now competing directly with their Model vendors and need all the goodwill from developers they can get.
AI Pricing is Still Hard
When building an AI product that relies on LLMs from OpenAI, Anthropic, et. al., costs will always be a challenge. Your costs do you not scale; as users use your product, your costs will go up linearly. This is a massive problem that breaks the typical SaaS business model (costs should go down at scale).
Cursor is at a size that it can negotiate deals with the Model Makers to keep prices predictable. Indeed, this was the only way they could make the Ultra plan work:
“Ultra is made possible by multi-year partnerships from OpenAI, Anthropic, Google, and xAI. Their support was instrumental in offering this volume of compute at a predictable price.”
But they are still beholden to the Model Makers for their pricing. Long-term, Finn Lobsien identified 3 ways Cursor could control their own destiny:
Orchestrate models so well that the experience is much better than any model vendor’s coding app.
Win on distribution with a UX/collaboration experience no one else can win.
Bolster its margins by developing its own models.
The “Tab” model is Cursor’s attempt of relying on its own model and creating an unlimited coding agent. They seem to be banking on this future. We’ll see if they are able to transition from App Maker to Model Maker.
Shout out: If you’re interested in AI pricing, follow Finn Lobsien and his company, Lago, on Substack:
A Few Good Links:
Here are a few more interesting stories right now:
Plus, a follow-up to yesterday’s story on Rainmaker:
Finally, a follow-up to Monday’s story on Drive Capital:
📚 One Page: Building Rocketships by Oji and Ezinne Udezue
Building Rocketships is a product management bible for high-growth companies.
Oji and Ezinne Udezue are both Chief Product Officers: Oji most recently at Typeform, Twitter, and Calendly; Ezinne at WP Engine and ProCore.
Pricing is one of the trickiest and most-important aspects of a product manager’s (or a product-focused founder’s) job. Accordingly, the book has an entire chapter dedicated to pricing strategy.
This page could have helped Cursor navigate its pricing communication challenges:
Making the Pricing Scheme Simple and Easy to Digest
The final principle is often undervalued. While your product is scaling, it’s important to use a pricing scheme that customers are both familiar with and that doesn’t offer obstacles to buying decisions.
The standard pricing model in SaaS today is the per-user subscription, but hybrid pricing schemes are becoming newly popular—the challenge is presenting the mixed pricing scheme in a simple and obvious way.
While you are scaling, focus on the ability of your target customer base to easily understand your pricing, keeping a keen eye out for objections. For example, foisting usage-based pricing on customers who are accustomed to subscription-based pricing can be counterproductive if the pricing itself adds friction to the buying process. Only introduce these changes where the benefit is very high—for example, usage-based pricing can often mean significantly lower per-customer revenue in the short term, but higher conversion, a potentially good tradeoff.
If your product is often paired with other software products (for example, as part of a marketing tech stack), it’s advantageous to be priced similarly to those products. If your pricing scheme is different or an outlier, it may be harder for your intended customers to pay for your tool differently, and may push them to look for alternatives. Basically, don’t charge in Bitcoin when everyone around you is charging dollars, unless there is a darned good reason!
One of the most interesting things we have seen as technology companies experiment with pricing is the one-price company—startups will offer a singular price (paired with a free tier). This makes sense in many ways: There are no confusing price tiers to choose from. It’s a simple, single, accessible price point. In reality this cleverly masks a startup’s inability to invest in many features that can be price differentiated early on. So it’s win-win for the startup and customers alike. Customers get a great accessible price and fewer decisions to make about products they want to use. Startups get to NOT worry about the intricacies of pricing beyond their current stage and have time to expand their offerings (and set new price points) as they grow and expand.
Building Rocketships is a Damn Gravity title. Want to read more? Get it here:
❓ One Question: AI productivity
Today’s question is more about scratching my own curiosity:
In what ways are you using AI to be more productive in life and work?
and as a bonus:
Is AI helping you create better work, or just the same work, faster?
Don’t just ponder this question… reply in comments and let’s chat.
🗳️ Wrap Up and Feedback
That’s it for today’s Daily Grind! I’d love to hear what you think of this new format. Take the poll below and reply with any feedback or ideas.