An update on the Cursor article

In September 2024 we published a short piece about the afternoon we first used Cursor with GPT-3.5 and decided the cost of building software had changed for good. It is a rare chance to check a prediction cleanly.

What has happened since

When we wrote it, Anysphere, the company behind Cursor, had raised an $8 million seed round and had around thirty thousand daily users at the end of 2023. The month before the article went out, it raised a $60 million Series A. By the end of 2024 it had raised a further $105 million at a valuation of around $2.6 billion. In May 2025 it raised $900 million at $9 billion. In November 2025 a Series D priced it at $29.3 billion. Annualised revenue was reported at around $1 billion during 2025 and had roughly doubled by 2026.

On 16 June 2026 SpaceX exercised an option to buy the company for $60 billion in stock. The deal closed on 14 August, and Cursor now sits inside a new SpaceXAI division. It is reported to be the largest acquisition of a startup on record.

The investment I did not make

I wish I had taken everything I had and put it into Cursor at the point of writing that article.

It was a private company and we were a newly founded studio in Leeds, so the door was not open to us, and I am aware that the wish is a cheap one to make from here. It still stands. We had seen the thing clearly enough to build a business on it and to say so in public, and we never thought of the tool itself as the asset.

What we got right

The constraint moved. Implementation stopped being the expensive part of software, and judgement became the job. That is now how we hire, how we staff projects and why we call the people who run them product engineers rather than developers.

A small senior team with these tools can do what used to take a large one. Every project we have delivered since has been built by fewer people, in fewer weeks, than the same work would have taken in 2022, and the software has held up.

The models improved on schedule. GPT-3.5 made mistakes a junior engineer would make. The models we work with now make mistakes a good senior engineer would occasionally make, which is a different category of problem, and the reason the verification we described in the production grade article still matters.

What we missed

The speed of it. We expected the market to catch up over a period of years. Cursor went from a seed round to $60 billion in a little over three. The window in which a small studio has an edge from using these tools well has been shorter than we planned for, and the edge now comes from the judgement and the process around the tools, because the tools are available to everyone.

We also underestimated how much of the value would accrue to the tool rather than to the people using it. That is the part I think about most.

What it means for the people we work with

The founders and businesses we build for have gained the most from this. A first product now costs a fraction of what it did when we wrote the original article, and the argument we made in Spend the raise on selling is stronger than it was. The pressure on older software products described in the SaaSpocalypse piece is the same force, seen from the other side.

The original article is unchanged, with a note at the top pointing here. We will do this again in another two years.