Devin just came to take your software job… will code for $8/hr

Devon it's a name that strikes fear Into the Heart of every software engineer and computer science student it's just one letter away from being the devil and about 8 months ago I made a video warning programmers that Devon was coming and coming to take your gerb because it's a fully automated Junior engineer that can build test and ship code better than 74.2% of humans well today I have the unfortunate responsibility to tell you that Devon finally came and now your manager can hire Devon for just $500 per month and fire all those unreliable code monkeys

who cost up to 100 times more usually you get what you pay for but in today's video we'll find out if Devon is the real deal or just the latest overhype chat GPT rapper in the AI SLO averse it is December 13th 2024 and you're watching the code report it's been a crazy week for AI mostly because every product manager out there needs to ship something before the end of Q4 to hit their kpis open AI released their generative video model Sora but then almost immediately turned off account creation likely because generating video is extremely

compute intensive meanwhile character AI is being sued by the parents of an autistic teenager because the chatbot told the kid to terminate the parents after they tried to limit her screen time then Google released Gemini 2.0 with a feature called Mariner to take control of your browser and a feature called jewels to write and ship your code an idea very similar to Devon when I first heard Deon was released I was terrified but was relieved when I heard it get stuck in Vim just like the rest of us then immediately after launch it had a

critical security incident where one of its Prime members exposed a vs code live share URL allowing anyone to use Devon's machine it was resolved quickly though because Devon can use Deon to fix itself it's technology that made me Deon curious so I went to the onboarding page to try it out but was shocked when they hit me with the $500 per month entry fee however if we look at the pricing it's based on something called Agent compute units and it all works out to about $8 per hour which is a great deal for a software

engineer but they're really vague about what Devon actually is there's no benchmarks no nutrition facts no scientific papers it's just the name Devon in Devon We Trust though because its company cognition Labs recently raised $175 million at a $2 billion valuation I am a very rich that puts a ton of pressure on a young startup with no Revenue in a highly competitive field but we do know that they have their own models using proprietary data that proprietary data likely be in your GitHub repos but apparently Devon's even better at writing code than open ai's 0

one that's based on their own internal trust me bro Benchmark which I'm sure is based on rigorous science and definitely has nothing to do with their recent product launch unfortunately I'm currently broke because I just paid $200 for chaty BT Pro but luckily I found someone who was stupid or I mean brave enough to buy it I paid the $500 a month to use Deon and that means I can put food on my family this Christmas in fact that person is not stupid at all it's the CEO of Builder iio Steve and I'd highly recommend

you subscribe to his channel if you're into software development the first weird thing to know about Devon is that it's not an AI like cursor and you can only interact with it via slack that's a weird design choice that wouldn't appeal to 99% of developers but remember this is a tool for non-programmers to replace programmers and they want to appeal to those big Enterprise workers who spend most of their time time in slack now in slack you can tag Devon and tell it to do something that you'd normally need to wait four weeks for a

developer to fail doing and that will spin up a workspace that includes a shell browser and editor where it can write run and end to end test the code and ultimately send a pull request to GitHub to push it to production and what's kind of cool about it is that you can follow along step by step in this planner along the way it'll create these things called knowledge entries to give it the proper context of everything your team is working on and I think we can trust Steve when he says that it writes pretty good

code but not perfect code like it was able to take an open source image generator run it and bring the result all the way back to slack that's cool and he also had it build a new feature into an existing app and it wrote pretty decent code although it had typical AI quirks like unnecessary packages but with Devon you can also go to GitHub and add comments to the PO request and it will then fix these issues although later Devon was asked to fix a bug in some existing code it did kind of fix it

but changed some other things in weird ways and when he asked for an explanation it hallucinated some nonsense most of this information is not actually true it appears to suffer from the same issue of every other AI tool where it works a lot better on established things like reactjs but then fails on Superior but less popular tools like spel despite being Rough Around the Edges Devon is still quite impressive nothing even close to this existed just a year ago but the main problem for me is the slack based workflow and it's kind of hilarious that

a company with Gen Z co-founders decided to build their product on Boomer Discord I for one can't wait to be made obsolete by Devon but if you're a developer who wants to build a future with AI One open source tool you need to know about is PGI created by time scale the sponsor of today's video time scale is a database built directly on top of postgress that supports time series data with extremely fast queries that use less compute but they recently released another amazing open source tool called pgai vectorizer it can transform your unstructured text

into semantically aware embeddings without the need for manual syncing or custom pipelines and that means building AI systems like rag or recommendation engines can be done easily at the database level not only does it simplify development but you also get incredible performance as a side effect you can self-host PG in any postgress database or better yet try it out on time scale Cloud using the link below this has been the code report thanks for watching and I will see you in the next one