Replit are listening

Agent & Assistant
, ,
1

I’ve just got off a great call with @FranciscoCM at Replit, where we discussed some of the current issues. Whilst I have been one of the first to say that these last 7 days have not been Replit’s finest (not least their silence), I feel reassured that they are listening and aware of what is going on, and are very much heads-down working on sorting things out.

Key issue

As per Amjad’s recent post on X, a core issue seems to be centered around agent 3 looking at agent 2 code and deciding to let rip and improve/change it. Once it was described to me, this kind of made sense.

As an analogy, any experienced developers reading this will probably agree that they’d rather start an app from scratch than try and get their head round another developer’s code! I personally come out in a cold sweat if I am made to look at someone elses code :blush:

If we ask agent 3 to build something from scratch, it will probably do a great job (has anyone done this yet???).

But,… if we ask agent 3 to work with agent 2’s code, then, like human developers, it is going a bit crazy - sending the architect and testing tools in to rip up the code and rewrite it in its own style.

I have summarised, but this scenario makes sense to me; I’d love to know if it resonates with others too?

Agent personalities

I also commented to Francisco that I have found agent 3’s “personality” and style of working very different to agent 2’s. In much the same way as ChatGPT 5 felt very different to ChatGPT 4-o - and was equally slammed on social media.

It feels odd to talk about agent personalities, and maybe that is the wrong word. But something changes with these big AI upgrades, which means we have to alter how we prompt the agents and modify our workflows to suite different kinds of agent responses.

Tell me if this makes sense to you? (or am I just going mad :rofl:)

Please give your feedback to Replit

I feel fairly confident that we will see and hear much more of the Replit team over time.

But for now, please keep adding a summary of your issues to my mega-post, as Francisco is going to ensure it is passed to the right internal teams for review: Summary of Replit issues - ALL MEMBERS, PLEASE CONTRIBUTE ASAP - Bugs - Replit Community Forum

That banned word which is the opposite of “long”

Totally unrelated to above, but Francisco is going to fix this platform to ensure we can use “that word”!! :rofl::rofl:

5 Likes
2

I told you Francisco was a good guy, Steve. Glad you had the call with him.

4 Likes
3

Thanks for sharing. I’m not seeing better results in the new app built from scratch with agent 3 compared to the one I started with Agent 2. The agent in the new app is also slow, disorganized, and sometimes not following my instructions.

In some ways it’s improved, but seeing agnet working 15 minutes and $8 only to discover the new feature doesn’t work is frustrating. That’s 15 minutes × 4 = 1 hour and $32. It’s not always that bad—there are some lighter moments—but spending $100+ per day (×24 days = $2.4k per month) without seeing major progress is way too much for me personally :confused:

Over time, things will improve. From what I’ve observed and read, Replit is unlikely to change its pricing model significantly. They perceive the situation differently: they recognize that the agent is continuously improving, and they are aware of the cost of hiring a skilled human developer, which ranges from $50 to $100, averaging around $70 in the EU.

They seem to be saying: If you won’t pay $20/hour for a developer with an IQ over 140 (agent 3), you can hire someone to do the same work for $70/hour.

Yes, many people won’t pay that much, so they leave for cheaper solutions. It’s only a matter of time before other players do what Replit is doing. We’re already seeing OpenAI’s Codex, which looks promising.

Cheers!
PS: I look forward to their upcoming update and bug fixes. :waving_hand:

3 Likes
4

It’s pretty easy to find a highly qualified full-stack and even mobile app guru for $20-$30/hr on UpWork.

3 Likes
5

Never tried Upwork :smiley:

3 Likes
6

It’s true that if you’re in the US or UK and you hire a full stack dev down the street, it’ll cost you quite a bit. But there are hundreds of thousands of them looking for work all over the world in the $20-$30 range.

2 Likes
7

Got it, but my point was how I see Replit thinking about their pricing of agentic coding and the future of code general. we will see…

1 Like
8

Steve, I completely recognize what they’re describing. And as you say, it’s not okay to have to mess around in someone else’s work. I’m allergic to that, and many others feel the same. So I totally get it. Let’s hope Replit comes up with an answer so we can code with a smile again. Thanks, you deserve some flowers. Great as well that you’re putting time into this. Awesome.

2 Likes
9

This is an important point @FranciscoCM - new apps started in agent 3 possibly aren’t showing good results either:

Anyone who thinks other AI dev platforms will remain cheap are a little naive. As soon as the VC investment runs out, they will turn the screws too. Hook early adopters in on subsidised prices, then reveal the true costs. The classic case of this was Uber, and every VC backed startups performs the same dance.

2 Likes
10

I wonder if anyone has given agent 3 some Lovable apps to play with - now that really would bring it out in a rash :rofl:

11

Thanks Eric :folded_hands:

1 Like
12

@Gipity-Steve we can now say short! Progress!

3 Likes
13

I raised this pricing issue couple of weeks back, even before they release Agent 3. They asked me to send a mail to support. I am afraid of even opening my account now because of the scary and opaque pricing. For a small player like me this clearly unaffordable. I am already exploring options Gemini CLI and Warp. To me Warp is really really good if you have some basic technical understanding and skills to use it.

14

Thanks for your effort on this Steve! Glad the communication channels are open.

I do have to say though based on this information, if Replit knew that Agent 3 was going to be disruptive to any code written by Agent 2 and that the entire user base was using Agent 2 then some fair warning and option to choose to continue with Agent 2 for existing projects would have been a good call.

Obviously no problem with all new projects being Agent 3 native as that is Replit’s direction.

On Agent personality yes its odd to say but Agent 3 feels totally different in all aspects of work and communication, maybe need to get used to it but time will tell.

2 Likes