Why I Skip Cursor for Dojo Workspace

By Fatemeh Rostami — Software Engineer

When my friend Val Neekman was building his AI agent, he told me he wanted a feminine touch on it. So I became one of his testers. Not the kind who opens a tool, clicks around for twenty minutes, and sends back a list of typos — the kind who moves in. I ran real workflows through it. I leaned on the rough edges on purpose. I told him honestly what wasn't working, what confused me, and what I expected from a tool like this before I'd trust it with actual work.

!When Fatima met Dojo

Some of that feedback was uncomfortable to give. Telling a friend that the thing he's been building for months doesn't behave the way people will expect is not a fun message to type. But that's the only kind of testing worth anything. Praise doesn't ship features.

A while later, he wrote something I haven't stopped thinking about:

A piece of you is in Dojo Workspace. Your thoughtful feedback helped make it better.

I've rarely been prouder of a product I didn't build.

The transcript that started it

Here's the thing that earned that sentence, and it wasn't a clever bug report. It was me using the tool for something it wasn't really meant for yet.

I had a video I needed to summarize. Not watch — summarize. I wanted the substance of it in text so I could skim it, quote it, and move on with my day. So I asked ChatGPT. Then I asked Claude. Neither of them could actually transcribe the file for me; I got polite explanations and suggestions to go find a transcription service.

So, half out of curiosity, I asked Dojo.

It transcribed the video. It reached for a Python library, did the work locally on my machine, and came back in a fair amount of time with a clean transcript I could actually use. Then it summarized it.

I told Val about it, expecting mild interest. Instead he treated it as a signal: a user had walked into his coding agent carrying a media problem, and the coding agent had quietly solved it. That wasn't a use case anyone had planned for. He built on it — hardened the transcription path, extended it, and kept going in that direction.

Today Dojo doesn't just transcribe. It inspects media files and codecs, pulls thumbnails, trims and converts, prepares subtitles, and searches long footage for the moment you're describing in plain language. And that whole branch of the product started, in some small part, with a woman who just wanted to avoid watching a video.

That's why the sentence landed the way it did. Not "you found bugs." You used it in a way I hadn't seen, and it changed what the product became.

So here's what it actually is

Dojo Workspace rests on a single conviction: AI should develop people, not replace them. Most AI tools sell you access to a model. Dojo sells you a better way to organize the work around it.

🥋 Dojo Solo gives you one focused lane for heads-down execution.

🥋 Dojo Duo puts a Sensei and a Grunt side by side — planning, research, peer review, and validation running in parallel, ideally with a different model in each lane so one can genuinely challenge the other rather than agreeing with itself. Instead of forcing everything through one enormous context window, Dojo keeps the work focused across coordinated lanes. Less context bloat. Better answers. Lower cost.

It mentors while it executes, explaining its reasoning as it goes, so you grow into a sharper builder instead of a button-pusher. It's local-first by design: agent work, browser automation, terminal control, and voice all run on your machine. No telemetry, no data collection. And you bring your own model provider.

That last set of choices is the part I'd push people to notice. Local-first and bring-your-own-model are not marketing features; they're expensive architectural commitments that make the business model harder. You only make them if you actually believe the user should stay in control.

Val's framing for the whole thing is a line borrowed from a cartoon chef: anyone can code. Not everyone will be great at it without practice — but the door shouldn't only open for people who started young, studied the right subject, or learned to enjoy error messages. A dojo isn't a vending machine for answers. It's a place you go to practice, with someone experienced close enough to keep frustration from turning into surrender.

And here's the part that surprised me most

It began as a tool for engineers and coders. But somewhere along the way, it outgrew that box.

The same engine that reads your codebase before touching it can now generate an image from a description, edit one while preserving the thing that actually matters in it, turn a still into cinematic motion, produce a voiceover in multiple languages, and render a flowchart or an architecture map instead of burying the explanation in another paragraph. A product photo becomes a promo video. A sketch becomes a photoreal concept and then a flyover. A three-hour recording becomes a searchable transcript.

Fast, too — genuinely faster than I expected the first several times.

Which means it isn't really a developer tool anymore. It's a tool for anyone who makes things.

I'm still using it and it's still teaching me

This isn't a story about a product I tried once. I started at a new company recently, on a multirepo project, and onboarding into an unfamiliar codebase spread across several repositories is one of the least glamorous, most disorienting things an engineer does. It's where you feel most junior no matter how many years you have.

Dojo is how I got my footing. I talked to it — actually talked, out loud, while my eyes stayed on the code. I asked it to draw me a flowchart of a request path I couldn't hold in my head. I asked for an architecture map across repos and got one I could point at in a meeting. Piece by piece, it helped me build a knowledge base over the whole system instead of a pile of half-understood files.

It's also disciplined about cost in a way I didn't expect. It manages context deliberately rather than letting the model burn through your tokens re-reading everything it already knows. And it ships with hundreds of preset skills you can pull in, plus the ability to add your own — so the workspace shapes itself around your project instead of the other way around.

Here's the part I'd rather not admit out loud: my company pays for a Cursor account, and there are days I don't open it. Dojo is faster, more efficient with tokens, and carries more tools in one place. When a free-to-me tool is the one I reach for over a paid one, that's not loyalty to a friend. That's just what's better on my desk.

Why I'm writing about it at all

I write mostly about my own engineering growth — the slow, unglamorous climb from mid-level to senior. This piece looks like a detour, but it isn't.

Being a serious tester of someone else's product turned out to be one of the fastest ways I've found to sharpen my own judgment. When you have to articulate why something feels wrong — not just that it does — you're forced to name the principle underneath your instinct. And the moment that mattered most wasn't me being clever. It was me being an ordinary user with a real problem, refusing to accept "I can't do that" as the end of the conversation.

If you're mid-career and looking for a way to grow that nobody will assign you: go be the most useful tester someone you respect has ever had. Bring your real work, not a test case. It costs you nothing but attention, and attention is the whole skill.

While much of the industry races to make people replaceable, someone chose to build the opposite — something that makes them more capable. That deserves attention.

If you build, write, design, or create in any form, give it a try. A piece of me is in it, and I think a piece of you could be, too.

Dojo Workspace is built by Val Neekman at Neekware. Read more about how it works at heydojo.ai. Written by Fatemeh Rostami, Software Engineer.