I applaud @jefftj86 for calling out in another topic the fact that sometimes (often!?), it isn’t the AI agent that is at fault, but us.

I’m not going to weigh in on the specifics of the example you were discussing Jeff. But…

It is time we stop sugar coating the fact that many people using AI dev tools are not technical and have zero development experience.

This is not a bad thing - I am basing my entire new startup around supporting such people. I love the fact that we can now bring software building to anybody with a keyboard.

But let’s be honest. Many of the issues we often see are because we’ve instructed the agent badly. Perhaps we need to be more honest with ourselves and ask, did agent just fail to fix the problem because I gave it the wrong prompt?

As us old-schoolers like to say: garbage in, garbage out.

1 Like

Such a clutch pro tip is to use something like WisprFlow or other AI dictation software and TALK to Replit. The amount of context you lose by lazily typing and getting frustrated is killing your project (and your wallet).

Dictate and watch how much more efficient Agent is with stream of consciousness context.

1 Like

:100: . Pre-requisites should be learning IFTTT, then unlearning all manners and adjectives.

@realfunnyeric that’s a novel idea. Are you dictating your prompt and then copy/paste into Replit?

I talk directly to Replit.

1 Like

I am a massive stream-of-consciousness fan @realfunnyeric.

I do it in many of my Linkedin posts. In fact, I’d say the posts that flop are always the ones I spent hours perfecting. Whereas the ones where I just chuck down a mad thought always do well.

My point being: stream of consciousness doesn’t have to be verbal. If you can learn to write that way, including when talking to an AI dev agent or even just to Claude/GPT, then you will always get a better dialogue and results.

I read some of the topics on here, and see the huge detailed prompts people are giving, or the extreme replit.md instructions, and am convinced they are why issues are arising.

It’s a hard thing to explain (sorry if I’ve not done it well). But vibe coding is about providing a combination of the right/clear instructions, but written (or spoken) in a much freer way than many are currently doing.

1 Like

Most people talk 4x faster than the type. So it’s natural to load better context with voice, resulting it better agent outcomes.

1 Like

Hey, you’ve not seen me type - there is literally smoke coming out from the keyboard right now :rofl:

1 Like

Same for me (175+) But it’s still more stream with voice for me. I backfill a lot more.

1 Like

as a Brit, most voice recognition tools are US-built and don’t understand me. They say “your accent is so cute,… but I don’t understand a word you just said” :rofl::rofl:

Have you tried Wispr? It’s nuts.

1 Like

I have found this to be a pretty intense learning curve (expensive one too) but I feel like I’ve had a crash course in infrastructure and full stack systems, especially when you get really ambitious. The issue is that these tools will give you a superiority complex if you think that it does more than it actually does. I have been working on website and servers for a better part of a decade so I have some foundation on how basic web server and web systems work, I have fixed every type of issue on wordpress, magento, joomla and other CMS systems. Most of the kids trying to build these huge apps (i have one in my family) want to create the next level of banking, understanding nothing about what makes complex systems run. I have rebuilt my diamond system 3 times which means i probably wasted 2-3k just in R&D, I see people crying over $50 in wasted credits when they just prompt:

Make me an app with blockchain and secuirty to revolutionize the banking system. Do it.

then they come on reddit or here to yell at everyone on why their vision is costing way more money than they expected or why its crashing, or why its not working correctly.

2 Likes

Interestingly enough, I’ve built a pretty cool blockchain-based app using smart contracts, all within Replit.

1 Like

Yeah but you’re actually pretty well versed in this stuff, I believe you could build that for sure. The non techy people are the ones that are having the biggest issues deep into the process.

1 Like

100% agree

1 Like

I wonder if Replit can maintain its position in the market of serving everyone - experienced developers and non-techies alike? It is the reason I went for it as my platform of choice.

I know Amjad’s big line is “creating a million more developers [or was it a billion?]”. But there are a few little stirrings in the upper atmosphere that make me feel it will end up as a powerful toy for existing/experienced developers.

I hope I am wrong. But the all things to all people in this game is not an easy play.

1 Like

I’m sure as it gets better it will accomplish this task for what everyone really needs it to do: make a small app that interfaces with small business software or data, automation and niche software for unserved industries that do not require sophisticated infrastructure that would be too small for any investment to pour into. Games and high level apps will be he domain of more experienced developers for more years i believe. I think we will get AI IDE’s within Unity similar to how Roblox is including in Roblox Studio that will start pushing that forward. Its possible i’m behind the curve and Unity is already on it, I would have to ask my kids to be honest; my son has been fiddling with Roblox Studio and said the AI agent is still not that great.

1 Like