Replit is an AI software development platform that allows users to make web applications via natural language prompts. It is sort of like ChatGPT for software. Although LLMs today can produce code, they can’t produce development environments to actually deploy software. This is where Replit differs from ChatGPT or Claude – not only does it write code, it deploys, hosts, and executes code. 

Replit falls into the general category of “AI for Software” as other tools like Cursor, Softgen, and Devin. These tools are very well known in the software industry but less discussed in marketing, for the obvious reason that most marketers are not writing code on a day to day basis. However, most marketers not only use software but in fact are basically human agents moving data, ideas, and messages between software platforms to achieve a given result. 

We decided to try building a Google Ads application with Replit as a proof of concept. Here’s what we decided to build: 

This is light web application which: 

• Consumes Google Ads performance by zip
• Maps each zip data point onto a geographical map of the USA
• Allows the user to also upload client locations to compare location vs. results
• Includes filters and metrics selectors to make the data accessible 

This is a pretty simple application, but actually quite useful for anyone buying media for multi-location businesses. It is also actually novel functionality which does not natively exist in Google Ads and is not easy to replicate with an LLM. It requires a real software app to do this. 

Here’s an example of the output of our test Replit app: 

This app actually works! The features include: 

• Color coding zip points based on performance
• Filtering each zip by metric
• Zooming in and out, moving the map around, downloading the map screenshot
• A table of data for reference
• Ability to easily visualize different metrics just by using a dropdown
• Plotting client locations on the map using Anthropic API to find latlong from address

It was built for a hard cost of about $30 in Replit tokens, and took about 7 hours of human work to prompt Replit to make it, debug it, and get it live. We feel this represents a rather monumental shift in the role of software in business. 

What Does this Mean for Marketers? 

In general the power of tools like Replit is good news for savvy modern marketers. Soon enough every smart marketer will have the ability to quickly produce their own simple web applications, and even small businesses will be able to have unique custom software to empower their teams to be more productive and raise their standards of work. 

For media buyers, we can easily see how we could continue to work on our geo analysis tool, by making it part of a larger set of tools doing things like analyzing search queries, automating reporting and analysis, using AI to analyze trends, and so on. 

Interestingly, Replit both *IS* AI and *USES* AI: in other words Replit itself is an AI platform, and the software it produces can use AI. In our mapping example above, we used Anthropic to convert addresses into latitude and longitude, enabling mapping addresses easily. This is a simple application of a very high-leverage principle. In other words, the advancement of Replit and similar tools is a dual advancement – the software itself is better because of AI, and the software is now almost self-generating again because of AI. 

The Experience of Using Replit

It’s worth noting that using Replit isn’t as simple as just writing prompts and getting perfect finished software right away. As we noted in our blog about ChatGPT Operator, even Replit is still somewhere in the “proof of concept” stage of product development. 

Replit often gets stuck on errors and has trouble fixing them. Of the ~7 hours it took to make our mapping tool, 6/7 were spent debugging, repeating prompts, passing error messages back and forth, and so on. Sometimes Replit gets stuck on irreparable errors and needs to be rolled back or given some hack on how to solve the problem. (For example it could not figure out a bug with the Anthropic API key and required me to give it a series of workaround prompts to fix.) 

I imagine that as our app becomes more complex, the need for a human software developer to intervene will grow. On the other hand, obviously the platform’s capabilities will grow as well. Today the experience of using Replit is simultaneously amazing and frustrating. I found myself actually getting mad at the Replit agent giving it prompts like “why are you still making the same mistake!” Hopefully it will forgive me should it ever become sentient. 

What About the Future of (Marketing) Software? 

We’re not the first to comment that with AI software generation becoming mainstream, the cost of software development begins to drop towards zero. Replit is so cheap to use it might as well be free. Of course, large scaled up applications will still have hosting costs, API costs, issues around scale and security, and so on. But it’s hard to look at something like Replit and wonder – in 5 years will anyone be paying for software? Or software developers? 

Now to be fair, Replit itself is software. I wonder how much AI code is in Replit itself – the whole idea starts to reach near absurdity when you imagine a developer using AI to make a software platform that uses AI to make other software platforms that also use AI. Did I use AI to write this post? Are you using AI to read it? A digression. 

In practical terms, for marketers I think we’ll simultaneously see: an explosion of cheap software tools broadly available from small self-published software entrepreneurs, a growing focus of most marketing/agency teams to build their own software, and pressure on incumbent pre-AI platforms. I think we’ll start to hear – “why should I pay $$$ for platform X when I could build the same thing in Replit for $200?” discussed more and more across companies of all sizes.