Self Learning/Improving Cloud Agents

Hi product hunt !

We’re currently building a multi-provider cloud alternative to Grokbot, and there’s one feature we keep debating internally.

What if your coding agents actually learned from you over time?

Not just keeping context for one session, but remembering things like:

  • how you like your code structured

  • the decisions you’ve made in previous projects

  • mistakes or corrections you’ve given them

  • your preferred stack, patterns, and workflows

  • the way you usually approach certain problems

So instead of starting from zero every time, the agent gradually becomes more adapted to you and hopefully more effective the more you use it.

The idea sounds great in theory, but I’m wondering how valuable this would actually be in practice.

Would a self-learning agent be a must-have feature for you, or just another nice-to-have gimmick?

And if you wanted an agent to remember something about you, what would actually be worth remembering?

68 views

Add a comment

Replies

Best

Code style and naming habits are probably the first things i would save.

 Yes i was thinking on that,

for example, when you tell him a guideline or you take him back because you want him to do something differently instead of doing that every time he learn by itseld

This is definitely something we need, as long as we have fine, tuned control over what it remembers. Long term memory can be amazing, but it can also hold onto bad habits or outdated patterns from a quick prototype, which can lead to issues when it tries to apply them to productiion code later on.
If you decide to create this, please include a Memory Management panel where we can easily review, edit, or delete specific learned behaviors for each workspace

Here are the top things we’d love to keep:

Code style and formatting rules that linters often overlook.
Specific workaround patterns for those niche thirdparty APIs we use all the time.
Team specific boilerplate setups to make switching between repositories a breeze.

I can see this saving a lot of repetitive instructions. If I keep saying do it this way the agent should eventually get it.

 Hmm okay thanks for the feedback !