What’s your cost on them? $20/30hr? (genuinely curious)
Mannn I hear you, I actually stopped coding several apps at the same time because the cost was getting too much… This is what my inbox looks like now…
Don’t get me wrong, it is still 100x cheaper and faster than ever possible before… but it is something i really have to keep in mind now… I have to focus my time/money on creating something I can monatize before messing around with another possible project…
Surely at this sort of cost you download VS studio and run a local ollama model or use open router or and LLM of your choosing or codex. If you are worried I can zoom call people to set up. Shocked so many people are paying thousands and not just swapping supplier. I really think replit should make a standalone app and a local LLM through mac silicon. They could still charge for deployment. Many of replit users need a cost effective solution. There are many.
Interesting to hear and experience. How can I be in touch?
Hey! Are you using Agent 3 for all the coding? If so, STOP right now and start using the Assistant. It’s way cheaper and faster, and it also only does what you tell it and doesn’t go around hunting other stuff to change.
Use agent to start the project and setup sverything, and then go to assistant to continue building piece by piece.
And 2 more tips:
- Start a new chat for every new feature or different problem you’re trying to fix. Also, if it start to run in circles, go to a new chat
- Chat with the assistant about your project telling it ‘don’t write any code’. This can be to describe what’s going on, offer ideas on what to do next or even tell you how to fix issues (aka debug whatever). That costs nothing, they only charge for code written.
You can also save some money if you can go edit the code. This can be made by copypasting into ChatGPT or others and asking how to do it (also good if assistant can’t solve the issue). After a couple of tries, you can ask Agent, but be aware because it can change unsolicited stuff by the way.
Get your idea here, but that’s a lot of trust in your clients. You need to make sure they are going to pay and not just take and run. Also there’s a high chance for them to think why can’t I just tell it what I need and have it fix something and it breaks the app.
This is similar to freelance work in copywriting/copy editing. You hand over the finalized version when the client hands over the cash.
yeah that’s a big chunk — $30 per hour adds up fast, and $360 in one day is definitely shocking. make sure you’re tracking hours so it doesn’t sneak up on you again
I think even with opus i’m hitting around $25 per hour on light work but yeah for complex stuff it can get up to the $35ish range, but these are huge code edits, thousands of lines of code major refactoring and feature development, after the main work is done, edits range around $6-$10 an hour at that point with medium autonomy
Thats about right for my usage too.
I spend about $500/day in credits. I’m proud of it honestly, the pace I’ve been moving at with Agent 3 is otherworldly.
With that said, Agent 4 seems to cost much less… but I feel like I’m being forced to work at a fraction of my former speed.
I’m not a fan of Agent 4 yet, and I’m used to spending hundreds every week also but now it tries to work autonomously on its own and the stop button doesn’t actually stop it. I had my Agent doing some easy updates - things agent 3 would do without issues, take 30 minutes and about 20 minutes in i tried to stop it to revise the instructions because it went off the rails, and it would NOT stop. I don’t like the new task system, it is kind of confusing- and the plan/build button should be a drop down again, its not obvious when its off or on, at minimum it should be more bold. This will cause me to take a break for at least 2-3 weeks in development until the bugs are worked out, again. I applaud Replit for trying to move fast in developing these systems and give us first access but, dang is it annoying when things don’t work and you can’t use the older system. We are moving into alot of us are already IN PRODUCTION with our software. I think the team may underestimate how big changes in how the workflows run change the way we instruct these systems, and now we have to re-learn how to get these systems to achieve updates or changes to existing code without trying to refactor everything or have it try to address things I didn’t ask it to do. Maybe I missed the memo but I’m down for at least 7-14 days until I see more positive info from the community, Im not burning cash for nothing
