Big fan of your videos. This tech has come a REALLY long way in a VERY short amount of time. Here's what I want to see next, or will build if it's not around in a month or so. I need the model to retain context about existing code, especially with regard to interdependent internal libs, code style including Prettier and ESLint instructions, and creating/editing/organizing files consistently with the code base that's there and in intelligent ways when adding new things that aren't mappable to a previous paradigm. For it to be production-worthy, it must also have some concept of the design patterns and UI kits available to it, and know how to operate within those confines. This especially includes conventions and components that aren't on the Internet at large, but instead have been built by a specific team and used in a proprietary way. Once we have each of these available to us, the options for larger projects and enterprise applications will open up quite substantially.
So far I've used it to generate two small apps. Web based inventory scanner to use on a tablet as a portable inventory scanning tool. Peptide dose calculator Total cost $0.35 VS Using straight up Opus to design a similar inventory scanner at $20.16 (this one used a backend and connected to the database, the other just made a CSV file on the backend) Nice!
@@jasonliu6321 It was a lengthy prompt detailing the form and the data format. It was a web page with JS, python and flask. At my request. Here is the prompt I used to create the Peptide calculator: I would like to make an app using HTML, CSS and JS. This app will be used to calculate the ratios of reconstitution fluid to product and how much to pipette to achieve a certain mcg dose of product. I want to be able to input any two of the values and have the app calculate the third. I also want to show how many doses are available in the reconstituted solution once the mcg per dose is known. The page should be modern, well spaced and in a darker theme with bright text and buttons. Buttons would be CALC and CLEAR. the name of the project can be dose_app It worked on the first try and looks quite nice.
I updated my Devika yesterday to use Sonnet 3.5 and the performance was greatly improved. Still not great but i think that’s a framework issue not the model. Will try maestro for sure
I wonder if we should make a distinction between “serial single agent at a time “ agentic and “multiple concurrent agents “agentic just to be clear about the categories of agentic approaches.
Can you show us a way of, instead of using sonet, we could use a local run llm? Like, my ollama container is on port 8087, how would I use it here? thank you!
Any idea what the "Do you want to add a text file?" does? Text file to describe the project or text file to include data etc.. ?? First test was a success. Thank you !!!
Can it be used to improve on current code base? One thing I have been struggle with this type of system is it is great to generate some POC but if you want to iteratively improve on it, it start to struggle to understand the context
Big fan of your videos. This tech has come a REALLY long way in a VERY short amount of time.
Here's what I want to see next, or will build if it's not around in a month or so. I need the model to retain context about existing code, especially with regard to interdependent internal libs, code style including Prettier and ESLint instructions, and creating/editing/organizing files consistently with the code base that's there and in intelligent ways when adding new things that aren't mappable to a previous paradigm.
For it to be production-worthy, it must also have some concept of the design patterns and UI kits available to it, and know how to operate within those confines. This especially includes conventions and components that aren't on the Internet at large, but instead have been built by a specific team and used in a proprietary way.
Once we have each of these available to us, the options for larger projects and enterprise applications will open up quite substantially.
Video resume: """It's pretty cool""
Thanks!
Thanks a lot.
So far I've used it to generate two small apps.
Web based inventory scanner to use on a tablet as a portable inventory scanning tool.
Peptide dose calculator
Total cost $0.35
VS
Using straight up Opus to design a similar inventory scanner at $20.16 (this one used a backend and connected to the database, the other just made a CSV file on the backend)
Nice!
what are your prompts to generate the inventory scanner? is it just one simple text or complicated description file?
@@jasonliu6321 It was a lengthy prompt detailing the form and the data format. It was a web page with JS, python and flask. At my request.
Here is the prompt I used to create the Peptide calculator:
I would like to make an app using HTML, CSS and JS. This app will be used to calculate the ratios of reconstitution fluid to product and how much to pipette to achieve a certain mcg dose of product. I want to be able to input any two of the values and have the app calculate the third. I also want to show how many doses are available in the reconstituted solution once the mcg per dose is known. The page should be modern, well spaced and in a darker theme with bright text and buttons. Buttons would be CALC and CLEAR. the name of the project can be dose_app
It worked on the first try and looks quite nice.
can you do a Maestro + DeepSeekV2 Coder Video like this ?
🎉🎉🎉🎉
man we want it badly specially its api is free
I also want "Maestro + DeepSeekV2 Coder Video".. +1 👍
@@jawwadnoaman8508 how you got deppsekv2 api for free?
I updated my Devika yesterday to use Sonnet 3.5 and the performance was greatly improved. Still not great but i think that’s a framework issue not the model. Will try maestro for sure
I wonder if we should make a distinction between “serial single agent at a time “ agentic and “multiple concurrent agents “agentic just to be clear about the categories of agentic approaches.
I think thats a good call if it isn't already done.
Does it work with local models like qwen2 or deepseekcoderv2 ? If so is it actually the best llm agent framework?
Can you show us a way of, instead of using sonet, we could use a local run llm? Like, my ollama container is on port 8087, how would I use it here? thank you!
There's an Ollama file which you can configure and use.
@@AICodeKing king
pretty cool video. well done king
I noticed most of the coding agents are doing small projects can we use it to create medium to large size business apps?
Yes
Hey someone know if its possible merge MAESTRO with MICROAGENT concept based in generate software with tests??
Can i create simple games to my phone without ads/micro translations? Then worth it.
Any idea what the "Do you want to add a text file?" does? Text file to describe the project or text file to include data etc.. ??
First test was a success. Thank you !!!
I believe it's for a bigger prompt if you want to give or for a context.
That's pretty wild!
Great video. Curious to know the cost for each app you generated.
How ‘cool’ was that 😊
Can it be used to improve on current code base? One thing I have been struggle with this type of system is it is great to generate some POC but if you want to iteratively improve on it, it start to struggle to understand the context
Recently I have covered Clsude Engineer. It can do that
What is the difference between Claude engineer and maestro
Maestro uses agents. Claude Engineer doesn't
@@AICodeKing how about from the end results what is the difference
So am i gonna be unenployed soon?😢
Start building AI so AI can't replace you. (Big Brain 🧠 Time)
Pretty cool.
"""It's pretty cool""
Good
Is this for real?
Yes
Gptcall app creator beast with sonnet
wow
Damnnn
I stile like aider
Thanks!
Thanks.
Thanks!
Thanks!