The biggest lesson I have learned building AI products this year: the model is not the differentiator. Control is.

The products that win are the ones that let users dial AI autonomy up or down in real time. Cursor is the best example of this. Give users control and you earn trust. That is how AI stops being a toy and becomes a partner.

At Canvas, this is not theoretical. Sketch sends you a Slack message with the content it drafted, a proposed email, or a suggested reply. You approve with a checkmark or ask for changes. The AI never acts on your behalf without your explicit go-ahead. That approval step is what turns an unreliable tool into a trusted teammate.

We explore this principle deeply in our piece on why human-in-the-loop is a philosophy, not a feature.

The Cursor 2.0 Strategy: A Risky Second Front

Speaking of Cursor, they just released 2.0. I absolutely love the product and have huge respect for the founders. But I am genuinely worried about their new strategy. Their bold move? Releasing their own model.

This feels a little like when Germany was fighting WWII. They were doing well against France and Britain, and then decided to open a second front against Russia, despite everyone warning them about the winter. This is what Cursor is doing in the AI war.

The Two-Front Cash Burn Problem

Front 1 (Inference): AI-first products are already bleeding money on inference costs, using models from OpenAI and Anthropic. It is not a profitable business today.

Front 2 (Training): Cursor's answer is to open a new front by training their own models. This adds a second massive cash burn (training costs) to their existing one.

When you are bleeding money on one front, you do not open a new business that also bleeds money.

The problem is that training models is an absolute commodity. It is a race to the bottom. Yes, it can be a moat, but it is a moat you have to fill with an ocean of cash.

I am saying this as a huge fan. The UX is on point, and I love the product. But this feels like a crazy, crazy move. It just shows how brutal it is to build a sustainable AI business right now. You are forced to choose between two different ways to bleed money.

As we discussed in the era of bigger-is-better AI being over, the value is shifting from raw model power to smart application of existing models. Cursor's bet goes against that trend.

Frequently Asked Questions

Why is user control more important than the AI model?

Because trust drives adoption. Users who can control how much autonomy the AI has (from fully manual to highly automated) feel safe using it daily. A brilliant model that acts unpredictably creates anxiety, not productivity.

Is Cursor building its own AI model a good strategy?

It is a high-risk move. Training models requires enormous capital and competes directly with well-funded labs like OpenAI and Anthropic. For an application-layer company, the smarter play is typically to differentiate through UX and user control rather than entering the model training arms race.