AI agents can prototype apps… But shipping real software takes hours of testing, debugging, and refactoring. Agent 3 is 10× more autonomous — it keeps going where others get stuck. The “Full Self-Driving” moment of software.
Automated Testing. While building your app, Agent 3 will periodically test it using a browser, generating a report and fixing the issues it finds. Our proprietary testing system is 3x faster and 10x more cost effective than state-of-the-art Computer Use Models.
Longer Autonomous Runs. Agent 3 is 10x more autonomous than V2, capable of handling much more complex builds by detecting and fixing errors on its own. You can track the progress of your build with Live Monitoring on your phone, freeing you up to focus on other creative work.

Sep 10, 2025 · 3:36 PM UTC

One more thing. Agent 3 can now generate your own agents. Agents that can work with you in Slack or Telegram. Or run based on schedule or trigger based on webhook events. Enrich it with your own data sources, and build automations with apps like Notion, Linear, Dropbox, etc.
Replying to @amasad
I'm past "man goes to doctor" jokes now and starting to wonder whether you should be paying us royalties.
1
2
29
Replying to @amasad
It’s also very expensive. Tried it yesterday and it spend $41 on onr prompt and $9 on another prompt before it stopped due to limit. Also I had set limits to $25 and still it used little bit over $50?
1
10
Replying to @amasad
Next up 2000 minutes
11
Replying to @amasad
this is a silly graph
6
Replying to @amasad
How does it keep a handle on business logic over such long stretches?
2
Replying to @amasad
100% on the same path here. We have some agents running for the past 3 weeks on the Linux Kernel and making improvements, small very small model compared to large popular ones. Specialists should exist in SLM.
2
Replying to @amasad
if there would be a model that has a short term memory (context window) and long term memory (parameters) and a system to transfer from both you have agents that can run indefinitely. i want to try to make it but i dont have enough time
2
Replying to @amasad
Absolutely mind-boggling. What are the best resources to learn more about this? Your website/youtube/etc?
1
Replying to @amasad
Amazing job, congrats. High level, without giving anything proprietary away, what are the top 3-5 things that let it work autonomously for so long?
1
Replying to @amasad
this is like the nano banana model, but for vibe coding
1
Replying to @amasad
I would gladly volunteer to help improve this processed
1
Replying to @amasad
Nice job. The value gets better by the week.
Replying to @amasad
Super excited to try Agent 3 at @3RD_AI_
Replying to @amasad
any plans to expose an api version of this?
Replying to @amasad
Congrats! Close to Waymo-length agent runs.
This tweet is unavailable
Replying to @amasad
Wait, what?
Replying to @amasad
20x more autonomy, 20x costlier
Replying to @amasad
Wow that’s is one crazy exponential it kind of mirrors the METR graph in a real world product context
Replying to @amasad
Agent 42
Replying to @amasad
I don't buy it
Replying to @amasad
The issue for small developers with the current Replit Agent is cost. It can run for an hour and get a lot done, but if that hour costs 60 dollars like it does atm, no one will use it. As of yet I haven't seen the 10x cost reduction.
Replying to @amasad
So it is text generation all the way down?. That's how you create value?
Replying to @amasad
This is both incredible and scary. CC: @DMattin
Replying to @amasad
You should implement this for your customer service. I can't get my account active after going back and forth with customer support for a week. I updated my payment information and still nothing. I need replit for my business help