What makes people actually stick with an AI coding agent?

We've been thinking less about how to get someone to try an AI coding agent and more about what makes them come back.

Getting someone to run a few prompts is easy. The harder part is making the tool useful enough that it becomes part of how they actually build.

With Clixad, we're trying to keep the workflow as close as possible to using a normal coding agent: open the terminal, give it a task, let it inspect the project, change files, run commands, test things and fix errors.

At the same time, we're trying to remove some of the friction around access. No own API key, no required monthly subscription, free credits, and additional credits through advertiser-funded offers, forms and surveys.

We recently made our five cheapest models free to use within daily limits as well, so occasional users can get quite a bit done without spending credits.

One thing we're still learning is where the sweet spot is between giving people enough free access to make the product genuinely useful and keeping the inference economics sustainable.

For founders building AI tools, what has actually made users come back to your product rather than just trying it once?