Micro Agent, a New AI Coding Agent, in 60 Seconds

Learn more at: www.builder.io/blog/micro-agent
#javascript #typescript #react #figma

[@michaeltse321]

Part of the motivation to code and learn to code is the feeling of solving a problem. this takes away the joy of coding - lol - but brilliant idea!

[@AlexAegisOfficial]

so its tdd for ai btw this is what I mean when I say LLM's are low level components, they arent bad, we're just not building actual tools with it

[@msschubi]

So, when the test is shit your code will be shit as well 😂

[@steelydanfan321]

I think you are misunderstanding

[@Qthuluu]

Still a valid point though. There is no guarantee the test is a good test.

[@msschubi]

@@steelydanfan321 I don't think so. Example: the ai creates a test that's always true...

[@Steve8708]

you can write your own tests too and just have micro agent write the code that passes the tests

[@msschubi]

@@Steve8708 so, even the ki is better with tdd :)

[@cloud-native-corner]

Cool idea, it would be good to have something like TDD agent that helps you iterate on tests and keeps on adding functionality.

[@khaledsanny4817]

You want to lose your job faster 😂

[@brazeshtamang4362]

Writing good tests means you understand what you need. But our job is usually spent thinking what you need. Sure AI can perfectly define an email validator but not all problems are not well defined like that.

[@geekwithabs]

Love the concept. 🎉

[@Aubstract]

So now I have to sit there debugging/rewriting the tests instead of the code 👍

[@Steve8708]

Or just write the test yourself manually and then set micro agent at solving it - works just as well

[@nickwoodward819]

@@Steve8708 "works just as well". I think that's his point

[@agentstona]

@@nickwoodward819 no it doesnt because its wasting TIME and effort .. by doing redundant work ... THE ifrst method of generating a code and than debugging it is if ONLY when the code doesnt work as expected but THIS guys trash method is always start by debugging unit test for every single code which is WASTE OF TIME and waste of effort .... He is trying to create a solution for a problem that does not exist resulting in creating a problem ....

[@nickwoodward819]

@@agentstona speaking of wasting TIME - you might want to re-read my comment. It can work just as well and be trash.

[@agentstona]

@@nickwoodward819 no come on man its stupid to Start with ALWAYS DEBUG than Write CODE .. its redundant and unnecessary ... DEBUGGING is supposed to be done when the program needs to be Debugged obviously after writing the code ... Common sense .... not wrting tests for a program that does not exist and trying to debug the test thats dubm.

[@trejohnson7677]

Damn assert based AI programming

[@abiolaolayinka2023]

Thanks

[@Iron_spider99]

Tell me you never used ai to create tests without telling me you never used ai to create tests.

[@oszi7058]

as he said in another comment you can write your own tests and give it to the ai

[@gabrielpauna62]

​@oszi7058 then you just writing code again

[@quinnherden]

​@@gabrielpauna62You are writing the tests, not all of the code

[@scottbotgo4218]

This is really cool dude

[@whebon7266]

So the AI is doing TDD

[@palashjyotiborah9888]

Be careful about the api costs. 😮😮

[@taheralipatrawala7300]

there's a logical interpretation gap between human problems and what AI can innately understand and solve. Once the gap is filled we may never face any problem with AI ever again. The workarounds like this one may be obsolete once the models get better

[@gaggioaxel]

This model has limited capabilities by structure. Not always predicting the next word (which is what this model does) is the correct solution to a problem. Some problems cannot be tackled by predicting a word. Just give it some math equations and you will see how this concept does not apply correctly. It just gives you the best results that please your ego. I’ve tried putting my university scores and it overestimated my average. But that’s also how the system was trained. To provide polite answers that are liked by the annotator rather than provide correct answers.

[@artwork-studios]

Can I use this for my web app?

[@Danzlh]

that’s cool. TDD with ai!

[@danielmontilla5915]

nice. now we need a tool to test the test to make sure the test is correct

[@minoakkyaw]

we might also need a tool to the tool to test the test to make sure the test is correct

[@Silver007xx]

Does this support uploading blocks of code? If I wanted the tool to review perhaps two files of code, would that be possible?

[@Steve8708]

Current doesn’t but have been thinking through the best way to support this

[@Silver007xx]

@@Steve8708 perhaps a prompt which opens a unique URL to drop files into some type of bucket?

[@mohammedmiftah7663]

If the code ia not scraping or consiming an external api... i just asked it to compile and test it itself. Sometimes it works😅

[@PythonicSerpent]

Does this work for other languages like c++ and java?

[@Steve8708]

Sure does!

[@upulwaruna.abeysinghe]

can i share d link to download

[@R_Y_Z_E_N]

Before , ai was used as SDE , now its an SDET 😂 . I liked you approach , but worst case it can take more time than normal . But best case it improves the code quality . 😊

[@faddoo]

So you have to pay for each test you do to OpenAI. You can create your own GTP for testing purpose.

[@user-iy9fr5td2f]

What about UI?

[@agentstona]

Nah this is not solving any problem its just trying to find a problem for your solution .... BY doing extra work .......

[@amaldev4150]

Now I have to debug LLM written tests!?

[@moises32123]

Just code your stuff 😂

[@warguy6474]

then ask ai to write tests that need to pass another test and vice versa

[@israelhilljr]

Nice, now i have to debug ai tests and ai code!

[@taulantus]

Very Cool! But it can generate bad tests, and debugging tests is annoying.

[@Steve8708]

Then just write the tests yourself, and have micro agent write the code that passes the test (it doesn’t need to generate the tests)

[@giftowercom]

Well we all know that tests don't cover the code 100%, and there could be 'overfitted' code that satisfies the test but is shit. Don't think writing tests is easier than writing acrual code. Anw interesting approach.

[@gabrielpauna62]

​@giftowercom yip actually it's much harder cause you work backwards

[@yuvrajchhatwani]

Its not free it need open ai key

[@ryujin2206]

Please dont tell me i will go to the website and itll say, basic pricing starts at $500 a month 😂

[@savire.ergheiz]

Skills issue 😂

[@_Zie_]

You're just moving the goalposts, now you have to write and debug tests. A broken clock is right twice a day. How about we as programmers stop making our projects by slapping layer upon layer of shitty code on top of eachother and actually just write decent programs, like people who actually understand computers and programming do.

[@agb2557]

Shouldn’t the human write the test?

[@Steve8708]

you can do that too if you want

[@agb2557]

Thanks, ​@@Steve8708

[@dievas_]

Learn to prompt then...