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