Cursor

Our verdict on Cursor

Cursor was Jordan's go-to IDE through most of 2025. In December 2025 he called it the most developed of all the editors he had installed, and the best experience when running several agents at once. By March 2026 our team had shifted most of its coding to Claude Code and OpenAI Codex, so read this as a review of a tool we used hard and then mostly left.

What we built with Cursor

• Ryz Score (July 2025): an AI resume scorer that rates every applicant on skills, experience, education and culture fit and pushes the score into our ATS. Jordan built most of the Vite, React, Tailwind and shadcn/ui front end in Cursor, with Supabase edge functions calling ChatGPT.
• Mate (July 2025): our internal company directory and culture app. Jordan scaffolded it in Bolt, took it out, and finished it in Cursor over roughly a weekend.
• Solopreneur finance dashboard for Collective (September 2025): first version in Bolt, then moved into Cursor, where Jordan used Codex to wire up the front end and APIs. The forecasts run on Gemini.
• Tix LAX (October 2025): a Mapbox map of Los Angeles parking tickets with an officer leaderboard. Jordan coded almost all of it with Cursor and Codex.
• Kalshi prediction market trading bot (January 2026): built with Opus 4.5, Cursor and Claude Code together. It won 8 of 12 trades and still lost about $7 to fees.

I think I built the first first version in Bolt just, like, really, really quickly just to, like, get something up.

— Jordan Metzner, Episode 14

How we use Cursor

1. Prototype somewhere faster, then move in. For Mate and the Collective dashboard, Jordan got a first version up in Bolt, then opened the project in Cursor for the real work (Episode 5, at 09:29).
2. Keep the data layer outside the editor. Bolt connects to Supabase directly, so the database and auth survive the move into Cursor untouched.
3. Pick a stack the agent knows well. Jordan's default was Vite and React with shadcn/ui and Tailwind, because they are responsive and look good with little effort.
4. Run a coding agent inside the IDE. On Tix LAX and the Collective dashboard, Jordan used Cursor as the editor and Codex to write and wire up the front end.
5. Run several agents in parallel. In Episode 24 Jordan said he was already running four agents at once, and that multi-agent work was where Cursor was strongest.
6. Use the browser and debug mode for UI work. This is August Kiles's tip from Episode 31: launch the built-in browser for UI tweaks, and let debug mode tell you what to click while it handles the logging.
7. Watch the bill. One of our developers used a new Cursor feature and left us with about a $2,000 bill (Episode 36, at 13:54). We now put guardrails on AI spend.

Where Cursor falls short

• Zero switching costs. Jordan made this point in September 2025: Cursor grew fast because switching is free, and it can lose users just as fast. That is roughly what happened on our team.
• Big, old codebases. Amar Goel of Bito said tools like Cursor are great for new projects but break down on large existing codebases, and asking them to review code finds shallower issues than a dedicated reviewer (Episode 23).
• An early enterprise pilot went badly. Arun Kalaiselvan said Arya Health ran a six-month pilot and removed Cursor. Anand Chandrasekaran added that this was 1.5 to 2 years earlier and Cursor is now very mature.
• Context, a year ago. In June 2026 Jordan remembered Cursor writing markdown notes with random, non-sequential numbers and no context. He contrasted that with now, when he can leave an agent running for half a soccer game.
• Throttling on new models. The night Gemini 3 launched, Jordan could not get it to work in Cursor.

We actually found that they introduced more issues into our code base than they solved for.

— Arun Kalaiselvan, Episode 24

What our guests say

Guest opinions split. Iliya Valchanov said Juma's whole team moved from Cursor to Claude Code and then canceled Cursor. Karim Malhas built Breeze with Cursor and runs a competition inside his company over who uses it most. Anis Bennaceur found in 2025 that his top and bottom engineers spent the most on Cursor, with nobody in the middle. Realm, in Episode 47, still uses it.

The debug mode is really helpful cause it'll tell you, okay, go and click on this, and then it automates the logging and everything.

— August Kiles, Episode 31

Cursor compared

• Claude Code vs Cursor: our team and Juma moved to Claude Code; August Kiles says the model matters more than the tool.
• Cursor vs Google Antigravity: Cursor is more polished; Antigravity is free with some cool one-off features.
• Cursor vs Lovable: Hooman Radfar's founder friends set up ideas in Lovable and move to Cursor for production.

Episodes featuring Cursor

• Episode 4: Ryz Score, mostly built in Cursor.
• Episode 5: Mate, from Bolt to Cursor in a weekend.
• Episode 15: Tix LAX, coded with Cursor and Codex.
• Episode 24: Jordan's IDE rundown and the debate on whether Cursor survives as models converge.
• Episode 35: Juma cancels Cursor; Jordan on our own move away.
• Episode 40: the SpaceX deal, and why enterprises are still adopting Cursor.

FAQ

Is Cursor worth it in 2026?

It depends on your team. In April 2026 Jordan said the most advanced AI teams had moved on from Cursor to Claude Code and Codex, but many enterprises and universities are only now adopting it. Guests like Emblem's August Kiles still use it for its built-in browser and debug mode.

How do you use Cursor with Bolt?

Jordan scaffolds a first version in Bolt, which connects to Supabase directly, then takes the app out of Bolt and finishes it in Cursor. He built Mate, an internal company directory, this way over about a weekend.

Cursor vs Claude Code: which is better?

Our team moved most of its work from Cursor to Claude Code, and Juma canceled Cursor because the gap was so big. August Kiles argues the model matters more than the tool, so Cursor with a newer Claude model can beat Claude Code with an older one.

What are the downsides of Cursor?

Arya Health ran a six-month enterprise pilot and found Cursor introduced more issues than it solved, though that was about 1.5 to 2 years before Episode 24. We also had one developer run up a roughly $2,000 bill with a new Cursor feature.