Cursor Was Never Bad. It Is Just Not Worth It Anymore

Cursor Was Never Bad. It Is Just Not Worth It Anymore

I remember when Cursor felt like the future. It was the first IDE that actually had an agent built in, not a plugin bolted on the side, not a chat window where you copy paste code back and forth. You typed what you wanted and it just went and did it, right inside your editor. Nobody else had that yet. That alone made it worth the hype at the time.

Cursor is also the thing that kicked off the whole vibe coding wave. Before Cursor, “AI helps you code” mostly meant autocomplete. Cursor made “just describe it and let the agent handle the rest” a normal way to build software. Whatever you think of vibe coding now, Cursor is the one that started that chaos.

Then came the pricing.

Cursor’s business was basically reselling access to other people’s models, OpenAI, Anthropic, whoever, wrapped in a flat monthly subscription. It worked like a gym membership. Charge everyone the same price and bet that most people will not actually use their full quota. The problem is a coding agent is not a treadmill nobody touches after January. People who use Cursor seriously use it every single day, and every request costs real tokens and real money. The math only works if usage stays low. The moment your product succeeds and people actually rely on it, the model breaks.

That is more or less what happened. Limits got tighter, pricing changed more than once, and using Cursor for real work started to feel like watching a meter run in the background.

And then the market caught up to it. Now there is Claude Code, there is Codex CLI, there is a pile of free or cheap harnesses doing the same “agent living in your terminal or editor” thing. Some are free, some just charge you for raw API usage without a gym membership markup on top. If you already pay for a Claude subscription or a ChatGPT subscription, an agent basically comes included. Paying Cursor on top of that stops making sense.

To be fair, I do not think Cursor is a bad product. The IDE is solid, the UX is genuinely good, and as a harness it does its job. It is just not good value for money anymore next to a Claude Code subscription or a Codex subscription. You end up paying twice, once for model access somewhere else, and again for Cursor sitting on top of it.

The one exception right now is Grok. Cursor paired with Grok 4.6 and Composer is actually a good deal. Some chatter I have seen lines up with what I think:

“You get tons of usage for Grok 4.6 and Composer which are genuinely good models. PLUS you get access to Grok bot. Also, the UX is good.”

“Cursor + Grok 4.6 is good and cheap”

That tracks. Right now that combo is one of the few real reasons to pick Cursor over just going straight to Claude Code or Codex.

I do not think it lasts though. My guess is Cursor slowly turns into what Codex already is to ChatGPT, or what Claude Code is to Claude, basically the coding front end for one model family instead of a neutral place that resells everyone’s models. Grok is the thing propping up Cursor’s value right now, so I would not be surprised if Cursor eventually just becomes a feature of the Grok subscription itself instead of a separate product you pay for on top.

Cursor did something that mattered. It proved agents inside the editor actually work, and it kicked off vibe coding as a real thing people do every day now. But being first does not mean staying best, and a gym membership pricing model was never going to survive an entire industry giving the same thing away for free or close to it.