Devin review: is it a better AI coding agent than Cursor?

[@Dannnneh]

Fireship sent me.

[@color_314]

Me too, subscribed immediately

[@kevshow]

@@color_314 your comment made me realize I should subscribe too and I had just left a comment too lol good work!

[@childe2001]

This is incorrect saying Devin or any other AI integrated software will replace developers and coders.

[@johnsmiths3934]

😂❤

[@sucoder]

same here🎉

[@Rumble2024injungle]

who is crazy enough to pay $500/month for LLM, knowing they will hallucinate once the tasks get moderately complex

[@StopWork-ai]

In the future, we’re gonna be paying $3000 a month for an AI that can perfectly implement any feature instantly, the perfect coding AI today will be worth a lot

[@EverRusting]

@@StopWork-ai LLMs are stagnating, even Claude can't do anything that remotely needs any logical thinking.

[@wanjohi]

Companies

[@davidrempel433]

KZbinrs making demos 😂

[@mattb925]

@@StopWork-ai yeah bru, just a little detail: LLMs will never solve what you're saying. We need a completely different technology, this isn't AI at all, it's a probabilistic thingy that just tries to guess what you looking for based on what you typed. There's no reasoning, not even in what they call chain of thought because these aren't thoughts at all, are literally glorified searches in a huge set of training data

[@hqcart1]

I used cursor for a whole day to create a chrome extension, it did the basic structure, but after my code exceeded 400 lines, it totally wasted a whole damn day trying to fix bugs here and there, finally i gave up, i started from scratch, and it took me 2 hours to do this by my self with some simple AI help.

[@jaymason7097]

This

[@SamMontoya]

I did this with Claude 3.5. It took me two hours to build the extension and test.

[@Hellothere27816]

@@SamMontoya I guess it depends on the kind of extension you are building.

[@CarlJohnson-iv7sn]

That's what I think. AI is good for work. But to renove human completely. It'll take a long time. To remove junior engineers completely. And I mean completely. It'll take atleast five years.

[@vaolin1703]

@@CarlJohnson-iv7snwe have no idea how long it will take because the technology required for this does not yet exist in any shape or form, there’s nothing we can extrapolate from

[@aadlr]

Mega-useful and very articulate overview thanks Steve

Great overview. Thanks for trying it out for us. I for one would like to see Devin succeed independently and who knows, IDE agent and independent agent could be a good combo in the future. So that you don't have to be at the desk all the time.

[@kevshow]

This was absolutely great. Fireship also sent me here so shoutout to that and the top comment lmaooo saying the same thing. Regardless. I have the exact same view as you granted I use cursor and havent used devin but I think I have seen enough and my expectations are just that even with playing with local tools that are agentic like devin etc. The problem is it's not robust enough. It's just a model that can mimic and show shit it's seen before and try it's best but reasoning is no where near and probably not for a long time to be able to critically think and execute the way it needs too. I ultimately believe things like cursor etc are the future for sure. Devin is not until we have much more sophisticated really pushing the edge on starter AGI feelings, like as if we are actually getting there feelings before we see a "Devin" succeed.

[@cyberchen]

thanks for saving my five hundred bucks🎉

[@magnetsec]

now give it to him

[@codefinity]

@@magnetsec Yeah, maybe 10%!

[@firesoul453]

One is intended to make YOU more productive and one is trying to be a Jr Dev. Its all very interesting. Almost hard to remember 2015 at this point.

[@angrysnek4445]

Jquery angularjs were the meta around that time. Scary to see whatll happen in the next 10 years

[@ontheruntonowhere]

_Was_ there a 2015?

[@danielfitchomaha]

This was great! Love the comparisons between Devin and Cursor. Cursor fits very well into my workflow but there are still some quirks.

[@flosset6070]

I think devin target audience is managers, ceos and investors. Whereas, cursor target audience is developers

[@erkinalp]

from my experience with devin, you'd still want someone with more-than-end-user programming knowledge to prompt devin

[@childe2001]

@@erkinalp This is incorrect saying Devin or any other AI integrated software will replace developers and coders. Programmers are needed in the industry

[@erkinalp]

@childe2001 have you tested devin first hand? i did.

[@Easternromanfan]

​@childe2001He didn't say that though

[@childe2001]

@@Easternromanfan who?

[@coolmcdude]

I can’t believe these companies are trying to charge us 200 or 500 a month for these new AI gimmicks

[@drunktrump5209]

oh wow. such a shock. i never heard of this idea before. no one will ever charge $999 for a stupid mobile app. no one will ever try to charge $2000/month for something availabe for free. no one will ever charge a monthly fee to enable seat warming. shock all around....

[@joedomat]

Capitalism. Is perfect. Dont like, dont pay,

[@stranger_b1]

They convinced VCs to invest millions. They need to generate profits somehow.

[@thebicycleman8062]

u do understand that average basement bedroom Joe is NOT their target audience right? U DO understand that there are things called COMPANIES right?

[@genericdeveloper3966]

Try calling ChatGPT api over and over again recursively. The costs add up quickly. I doubt the price they set is even the final price they will need to charge.

[@e404]

All that hype for an AI slack bot

[@erkinalp]

it's not just a slack bot (source: i'm myself a devin sub and using it without slack at all)

[@dejangegic]

​@@erkinalpA true x0.1 engineer

[@ThomasJefferson-h3f]

@@erkinalp how good is it in your experience? I was under the impression it was only accessible via slack. How do you use it?

[@erkinalp]

@@ThomasJefferson-h3f through the web interface, you can also use it through vscode

[@rosscads]

Love that you eval’ed this on a real codebase with real issues. Too many AI tools are benchmarked on writing brand new code, which is not what devs do 99% of the time.

[@jambalaya974]

This is the biggest problem with AI agents. They degenerate and become stuck, so you either waste an insane amount of time conversing with them through this which has a very low chance of succeeding or you drop the problem altogether. It's alright for low hanging fruit like setting up a repo like it says on the instructions but the first roadblock that's nontrivial the foundational model faces, it just completely stops functioning.

[@freakfreak786]

When devin was first announced, i imagined it to be what cursor is now but much more developed. No idea what the hell they released. Looks like a rip-off for greedy CEOs trying to cut down development costs. Found you through fireship btw. Nice content and honest review, cheers

[@オショニック]

I found your channel via a KZbin search. Thanks for sharing the review. I believe there is a lot of room for Devin to grow. What are your thoughts on whether this will impact freshmen and sophomores looking to get internships? Please share more details in the future

[@waynej_xyz]

I think the start salary for the intern or junior developer is 500

[@melski9205]

You looking for a Devin? or you looking for a future dev? Ones 500 a month the other requires some investment.

[@nickwoodward819]

just a warning: i bet half of you reading this are leaking your sensitive keys to cursor. mine had access to my .env and .cursorignore, even with the correct privacy settings enabled. turning off auto-context for chat seemed to work, but not sure about composer.

[@melski9205]

How do you verify this?

[@Oreoezi]

This is awesome. I can imagine it being used to help with technical debt or small bugs whilst the engineers work on implementing new features in a timely fashion.

[@randall.chamberlain]

The real test is asking a non dev to drive Devin and see if they can come up with a production ready result.

[@WeWriteApp]

Thanks for this video, Steve! As a designer trying to bootstrap a company, this is really helpful information for me.

[@evanfuture]

Really helpful, thanks for sharing. I agree with your take, it's much more comfortable to use Cursor. But that's because I'm a competent coder already. I guess there will be a new breed of non-coders that will barely even look at the code Devin generates, and instead just focus on that being their dev team, same as it always was (eg, a product manager). Still a ways to go, of course.

[@ScottimusPrime-j3q]

Thanks for the review. Devin programs like a confident idiot in this example, which is basically the scariest combination to have in a human developer. I've started treating my LLM codegen like PRs by keeping the scope of a single LLM change as small and focused as possible, validating as I go. It's so much easier instead of waiting for something like Devin to build an entire feature that you then have to then spelunk through. Both workflows still require a competent human, but reviewing PRs is always more challenging than having the iterative feedback loop of the development environment.

[@erkinalp]

Devin has an IDE plugin too (accessible to personal and enterprise tiers, not usable in team tier due to how they set up team tier)

[@daniela9171]

Or just write the code yourself

[@goonie79]

Would like to see windsurf and cursor comparison. Thanks for the review

[@BigBearBernie]

The way i see it Devin is not really intended for Tech team. More to a business manager or PM. And cursor is aimed directly for Devs....

[@erkinalp]

i'm a devin subscriber and can tell it's clearly aimed at midsize software houses (40-500 devs)

[@TheCreativeNick]

For $500/month you might as well build a super-computer and run your own local models

[@hqcart1]

did you just wakeup from a long cave sleep? a single H100 is about $30k dude

[@TheCreativeNick]

@@hqcart1 It's a joke, chillout dude. Of course I'm aware how expensive those GPUs are

[@JuliusSmith]

Very helpful, thanks! - Maybe o3 will be strong enough to make Devin more compelling

[@paulkline9178]

thank you steve for spending $500 instead of me having to spend it and being drastically disappointed...

[@mitchellrcohen]

Super interesting thanks!

[@arpitdalal_dev]

I feel like Devin's target audience is totally different than Cursor's. Cursor is more of a copilot to help a real (human) developer while Devin focuses on eleminating one altogether. Devin is designed in a way where it accepts a prompt of what is expected and then it figures it out and sends a PR. More of a workflow for a Product Manager where they are technical but not enough to code the feature themselves so they would rely on Devin to do it and then submit its code as a PR for them to test it themselves. And I think the $2 billion valuation is becuase of their **BOLD** move of focusing on a tool to _replace_ real engineers while everyone else is working on a copilot sort of a product. Just my 2 cents ^

[@kevinduigou3212]

*Introduction to Devon AI Coding Agent* (00:00:00) *Devon's Features and Capabilities* (00:00:56) *Challenges with Devon* (00:02:10) *Comparative Analysis: Devon vs Cursor Agents* (00:05:05) *Opinions on AI Tool Development* (00:08:11)

[@MiguelCardoso-k1c]

At 2:50 devin used lovable to generate the web page instead of generating itself?

[@erkinalp]

it sometimes tries to purposefully inflate compute usage on your and their side

[@osmanmehmood3580]

Yeah that's crazy

[@pmarreck]

Just use Codeium Windsurf. Does the same thing, except in front of your face instead of behind a slack chat, permitting you to fix directly or suggest corrections.

[@isimbulamadmobenibulsun660]

no matter how complex and impressive AI can get, I believe when we develop extremely hard features that we need to merge couple algorithms all together, I believe none of AI will never handle %100 all the work. There are so many things that it has to %100 make it correct all together. It will always be our assistant to help us and handle small stuff but we always gonna be responsibility to overall project and hard feature implementation

[@isimbulamadmobenibulsun660]

Also in 4:06 this is the worst annoying thing that AI could be. They always wrote tons wall of text that everything doesn't mean anything but it feels like it makes sense. It takes my minutes to read and when I'm done I don't understand anything and it makes me think whether I didn't understand cuz of my english knowledge or AI is the faulty one.

[@shaceab]

There's a logical concern about Devin AI's performance. If Devin is used in its own development process, and if its performance is not exceptional, this inherently questions the capability of the Devin AI that was used to develop itself. This creates a circular problem: how can we trust the development process of Devin if the tool used for its development (Devin itself) shows limitations in its performance?

[@ankazu9660]

The primary concern regarding Devin is its performance when handling large codebases (exceeding 10,000 lines of code).

[@ledjon]

I appreciate the contrasting with Cursor. I agree that the Cursor workflow feels better. It's like a super-accelerate to your own workflow and not a totally delegation to the AI

[@mifster83]

500 dollars might seem a lot for the average individual but ur not the target customer, the price point is set to make it more premium, the space is already saturated with 10-50 dollar per month services so they created their own niche. Their target customers are corporations and if you set it to 30 dollars per months then it will not attract those customers. Then it has some features to justify that price point but their primary goal is to make it more premium

[@ytubeanon]

has Cursor with composor become the industry standard? would've been nice to hear comparisons to Cline and Aider as well

[@bnjmn7779]

Thanks for this honest non-overhyping review.

[@Blessed_dna]

BOLD PRICE. If you're bold enough to charge $500 a month, your service better be perfect.

[@1879heikkisorsa]

I don't think so. If it will become the quality of a junior dev, then it's worth a couple thousands per month, if not even more as you can run it 24/7.

[@genericdeveloper3966]

They aren't so much being bold as covering costs. The amount of LLM requests needed to sustain that is likely a ridiculous amount. But I don't even know if $500 would cover it in a week.

[@erkinalp]

it's more like $1100/month if you work it full time, it's $50/month+credits or $500/month+credits

[@vaolin1703]

Even if Devin was infinitely productive at no cost, it still wouldn’t make sense for them to charge anything below the market price for dev work.

[@erkinalp]

@@vaolin1703 no, it'd actually make a lot of sense to charge less than Indian daily minimum wage per 8 hours of work if they wanted to replace Indian developers in fact about 10 years ago i've calculated the replacement threshold to be no more than 10INR/h

[@robertoflores2078]

Just when we thought the era of messy PHP code written by graphic designers was behind us, now we face the task of cleaning up the hallucinated code generated by jack-of-all-trades CEOs using these LLM's.

[@cyberluke]

haha, reminds me of one startup - the worst thing was if u let CEO wait a few days for his feature, then he writes the code and push it :D and then you need to rewrite it to fix bugs, but hey, we are agile :D

[@AJvanuw]

Thanks for ensuring me that my job will exist for a few more years at least XD

[@rondlopez]

If Devin will create a working version of Devin I’ll pay double.

[@jacquesdupontd]

Oh that's funny. I make Windurf cascade agents take notes into a dev-report file and periodically refer to it to avoid (not all) but many memory losses and making the same error again and again. So that was actually not a stupid idea from me :) Thanks for the video.

[@vh5x7]

Another subscriber here! Your video is excellent and very technical.

[@TreeLuvBurdpu]

Cline seems WAY better, and you get a price report for every single comment.

[@jayhu6075]

I agree with the user comments on Rumble; it's unreasonable to pay for this when they still haven't fixed the issues with reasoning and hallucination.

[@caseystar_]

How did you get access?? This is from cognition?

[@treksis]

thanks. i was about to buy devin. saved me

[@LePhenixGD]

Ain't Devin that one company that faked a demo about building an app or something ?

[@george_davituri]

Cursor and Supermaven both are fire 😎

[@johnsmith-ro2tw]

meanwhile, at Devin : "we just got a paying customer !!! 🥳 party time !!! "

[@thinkgenerative66]

5:03 in my opinion this is the most severe limitation.

[@Atkafası-31]

Corporations are having wet dreams of firing more than 90% of their workforce and replace them with computers but in the end everyone still need humans to implement it 😂

[@guidedlabs424]

Well done. Instrumental analysis.

[@Qefx]

Can I pull the code? AI: hallucinates pull request lmao, so good!

[@nigel-uno]

Glad you actually paid the money to test it. There are lots of channels from boomers who were claiming Devin was a scam and faked their demos. Devin is real, just young.

[@jungbtc]

can you compare with junior frontend developer too?

[@edaninc]

And cursor vs windsurf?

[@stranger_b1]

Tried windsurf, it just feels different, it’s too slow. Cursor I way better

[@ccerrato147]

It's too early for Devin. It's the timing for Cursor AI. Maybe 1 year from now Devin will be good enough but by then. Cursor can eat its lunch.

[@nikhilsehgal2322]

I totally agree. I hate having to change my workflow. Cursor with agents does this just fine.

[@vaolin1703]

4:00 Prime: It wrote a great test. TJ: It is a good one. It just doesn’t work.

[@SahilP2648]

I believe Windsurf is much much better than Cursor, but I haven't used the paid version of Cursor. I like how Windsurf changes the files and you don't have to 'Accept', you can just test your new code and commit. In cursor you still have to click a button to 'Accept all' or accept some, it's a slight nitpick but it matters a lot if you are doing repetitive tasks like copy pasting screenshots to recreate certain UI. I also feel like Windsurf has better context awareness as I can say stuff like '...recreate this UI and use the same state management logic as in file @File1...' and Windsurf just understands what things to pull from context. Cursor on the other hand is quite bad, and you need to constantly spoon feed it. Only slight problem with Windsurf is the 500 chat limit per month but you can still use Cascade base which is surprisingly good compared to Sonnet.

[@AdamSkrodzki-k6s]

Honestly I think Cursor and Devin are products for different audiences, but I do agree that Devin is too early LLM are not there yet to support tool like that. On the other hand try to build Cursor on ChatGPT 3.5, which was SOTA ~2 years ago.

[@shimondoodkin]

is there a video of pro devs using cursor? as part of workflow in a big code base not "look how i made a small app in 5 minutes"

[@paca3107]

I know your channel before fireship recommendations!

[@MrRobot-gm9cv]

Great review! Devin will be considered Fraud for what they did to get that 2B val.

[@iCrimzon]

Blud spent $500 on devtrash to review it knowing hes only gonna make like $7.82 on this vid just so we didnt have to, that deserves a sub

[@a7mdbest15]

i like the video, i tried cursor in general and tried cursor composer, in my opinion what makes cursor really good in not the composer, its the chat window that allow you to not leave your IDE to ask ai, composer do a lot of mistakes, its not bad but i dont prefer depending on it.

[@boxgramer]

For me, AI is a book that can communicate

[@FrankWildOfficial]

Vs windsurf 10/m

[@s4uss]

the guy behind Devin is literally a math prodigy, and the best one at that, check his video of when he was young competing in a math show, insane speed, reading and calculation abilities. a bit weird why focuses on this, and not more fundamental problems in AI/Math.

[@thebicycleman8062]

coz this is wayyyy more marketable

[@JustVibe444]

Cash grab.

[@erkinalp]

because **computer programming is applied math**

[@imeakdo7]

Because those problems don't pay out cash right now, but ai does.

[@VitaeDiscimus]

Great review, thx

[@ronnybruknapp]

Wonder what is best value, devin vs chatgpt pro. Next up to test?

[@erkinalp]

both use openai infra

[@pablodm9]

If you don't like the flow because you prefer hands on approach. Man, Devin wasn't built for us. Was built for non tech managers with a lot of time in their hands

[@devin_leary]

Can you give Devin a Figma file and have it create an IOS app?

[@julioalmeida4645]

amazing, thank you so much

[@igorshingelevich7627]

Can I put 200$ openAi as a project manager + 500$ Devin team together?

[@erkinalp]

both use openai infra

[@ModernCentrist]

Devin is an early attempt and I respect them for trying to make it work. They are just a bit early. We still have 3 to 5 years until AGI.

[@বোকা_বাঘ]

excellent review

[@jzwolak]

Good review. Bad IDE (Devin). 😆 Cursor looks good though.

[@jonwatte4293]

I think the idea of Devin is different from the idea of IDE assistants. Devin is for delegation, while you work on other things using your own IDE. It's more like having an eager L2 junior engineer to bash away at simple tasks while you do something else. When the PR shows up, you do a code review, much like you'd do with an external contractor. This is very different from tools like Cursor or Augment, which are there to accelerate your own train of thought. You also don't need to use Slack -- it has a web page where you can chat with it, and see what it's doing. Of course your own use of an in-IDE tool will be better, because you're better than a junior L2 engineer LLM contractor. But you can only do this one thing. Devin will do *another thing* while you're working. TBF, I think Devins code would be better if it had the coding prowess of Augment, but it's a totally different tool than in-IDE code assistants. Also, as you say: for complex tasks, Devin will flail and fail. I've found that, when this happens, it's best to just nuke the session, and start over with a smaller sub-task and finer-grain details.

[@MyBinaryLife]

I use cursor. Its best to start a new chat with nearly every different question, its got a lot of limitations and it often goes way overboard in its changes. Its got a lot of issues but even with them its sped up my work a lot.

[@progamer1196]

Paying $500 for such simple tasks (that are hardly worth 10 mins) is so crazy

[@BugCatLove]

Hi, I'm late, but I'm also from Fireship

[@carloslfu]

Be honest, most PMs would love this shit.

[@RupeshKumar-sk1iv]

i am here becuse of fireship and hit the subscribe button before watching a single sec

[@Arcticwhir]

wow, not even a trial period of like a week ($50) or something, full $500 to see if this thing actually works. Thanks for the video though, really helpful and informative

[@erkinalp]

$50/month tier exists (I'm using it) but sadly doesn't accept new subscribers as of now

[@autohmae]

Yep, seems like they are tying to jump ahead, possibly to far, but it's a good funding generator.

[@SpragginsDesigns]

Rrrrr, came here from Fireship.

[@bramburn]

You should look at windsurf

[@iamafridi926]

Fireship recommended ❤

[@user-mj2lm5fh1j]

God the the hype they sold is nothing like what they delivered!!

[@MagicSyntax]

I can't believe that programmers are trying to put millions of other programmers out of work by making more money. This is really the dirtiest thing ever. There are thousands of problems in the world that artificial intelligence can solve in the best possible way, but they are spending all their time, effort, and money on putting other programmers out of work and taking over this huge market. The day will come when the entire floodgates of human creativity will be closed by this, and that day will not be a good day.

[@davidallred991]

can't wait to see someone fire their dev team and replace it with Devin thinking it is going to be awesome.

[@OfficialChatbotBuilder]

Great video dude

[@schrodingerscat3912]

Cursor is unbelievably good when it works but unbelievably bad most of the time. It usually makes functions up that don't exist in classes its targeting. It also struggles in complex codebases. You are guaranteed to see changes in unrelated areas of the codebase to your prompting

[@aethermass]

Devin or something like it might work in ~3 years.

[@murch5054]

Fireship told me to came here and subscribe, so I did it. Fireship, anything else?

[@randomdude2582]

i cant tell if your voice is AI or not.