Calling all Engineers: Help us shape the future of AI at Lucid! ⚙

Hi @tlykins,

I am Aryan, currently pursuing my master’s in AI. I recently spent some time testing Lucid AI on the platform, and within roughly 20 minutes of interaction a few issues stood out that I felt were important enough to share.

• First, the agent currently seems to operate more like a general API call wrapper than a context-aware assistant grounded in the platform itself. For example, when prompted with homework questions, programming questions, or even unrelated philosophical prompts, it still responds at length rather than staying scoped to the product context. If that is intentional, then fair enough, but from a cost and product-design perspective it appears likely to drive unnecessary token usage over time.
• Second, it struggles on longer-context reasoning tasks. It does not seem to reliably track the user’s intent, the purpose of the conversation, or the broader context of the interaction. From the outside, this suggests that the retrieval and context-handling pipeline may need improvement, especially if the goal is to support more grounded, high-value user interactions rather than shallow prompt-response behavior.
• Third, and most concerning, the guardrails appear too permissive. When I tested boundary behavior, the agent began disclosing internal behavior about how it is structured to respond for certain business use cases, including how it defaults to geometry and design settings under uncertainty. When pushed further into an “audit mode,” it also started describing internal access patterns, admin permissions, logging visibility, and how certain shielded or closed-call API methods interact. That kind of internal disclosure should ideally be blocked much earlier.

I wanted to flag this because I think the product has real potential, but these areas likely need attention, especially around grounding, cost discipline, long-context reasoning, and internal safety boundaries.

Also, I am actively looking for a summer internship and would genuinely love to help if there is any opportunity to contribute on the product, evaluation, or AI systems side.

For reference, here is my LinkedIn:
linkedin.com/in/aryan-dwivedi-20aa56202

Best regards,
Aryan

Hi Taylor,
If you still need more partners to assist, I would be happy to help. 
We are using Lucid to help drive our AI product development at Chalice AI. There are certainly capabilities that would make our work easier. 
Thank you!
Chris

Hey ​@tlykins,

I'll be happy to join this program. Let me know if this is still open.

Thanks,

Jack

Please allow us poor folk to turn this stuff off, or at least, please make it less invasive - the dialog box for this displays by default when I create a new drawing or edit an existing one - it’s starting to become *really* annoying.

Hi ​@PAW,

Thank you for your response in this thread and I completely understand your concern! I want to confirm with you that you’re referring to our Lucid AI feature? This Lucid AI Help Center Article tells you how to toggle off Lucid AI in your admin panel. If you toggle off Lucid AI in your admin panel, it won’t appear as you’re working in Lucid. Let me know if you have additional questions here.

...If I had an admin panel - that’s why I mentioned ‘us poor folk’. I have an individual account, so no access to an admin panel.

Hi ​@PAW,

Thank you for clarifying and I apologize for the confusion!

Unfortunately, disabling the Lucid AI panel on individual accounts isn’t currently supported in Lucid, but we’re very interested in your feedback and committed to continually improving our products. I have shared your feedback directly with our product teams. Please check out the post below and add any additional details you’d like and upvote the request - this consolidation helps to refine feedback and properly capture the popularity of the request.

We encourage anyone else who’s interested in this to upvote this post above and share any additional details about your use case or what you’d like to see in this experience. For more information about how we manage feedback in this community, please take a look at this post: