My review of the Replit platform

Overall, Replit is awesome. As a platform, they have packaged so much into a singular experience. They have awesome dev features like an agent to help implement code, a general AI assistant with Claude that creates code diffs, and the ability to develop without AI. They have encapsulated console commands into stop and run buttons and pre-packaged many key components.

Additionally, they allowed reusable application components like Stripe, a database, cool UI features, and all of the build blocks needed to create an app. Overall the platform is exceptional.

Where I find that Replit lacks is the LLM’s ability to write code, plan, and the billing system on top of it. At the time of writing this, LLMs still generate mediocre code that you cannot build anything slightly complicated with. When you ask the LLM to fix something it often just deletes the part that does not work or mocks a response instead of getting it to work. This is probably why non-technical users are always so blown away by AI tools until they are capable of looking under the hood. Then again, its hard to take away from the drastic improvement in ease of use compared to learning a coding language, how to interact with a console, and eventually implementing something.

Due to these struggles, the pricing model of Replit is incredibly frustrating. At 25 cents per agent checkpoint and 5 cents per assistant-generated code, it can be extremely frustrating to spend credits on repetitions that do not work. Due to this, you can burn through the allocated credits quite easily.

Conclusion:

I think Replit is ready to benefit from the improved capabilities of AI models once they are released but should alter its pricing model to accommodate the uncertain success rate of LLMs. However, their platform is exceptional so kudos to the team!

If you enjoyed this free article and want to try replit, please consider using my referral link: https://replit.com/refer/colecrescas