Cluely: Multimodal AI Overlay

Summary

Roy founded Cluely after building Interview Coder, a tool to cheat on technical interviews.
Interview Coder went viral, led to conflict with Columbia and Amazon, and prompted Roy to leave Columbia.
Cluely is positioned as a multimodal AI desktop app that overlays the screen to assist users across tasks.
Company traction: launched one month ago, approaching $5M ARR, closed $5.3M seed round led by Abstract Ventures and Susa Ventures.
Roy frames AI adoption as inevitable and argues interfaces like translucent overlays are the future UX.

(immediate – Roy) Continue iterating Cluely UX to refine translucent overlay interaction.
(short term – Product/Engineering) Reduce latency and improve accuracy via model hosting and input caching.
(short term – ML Team) Develop custom system prompts and in-house evals based on usage analytics.
(medium term – Strategy) Build personalized, fine-tuned models per user to create a data moat.
(medium term – Growth/Marketing) Leverage controversial social posts to drive virality and product adoption.

Interview Coder: screen-and-audio AI overlay designed to answer live technical interview questions.
Translucent screen overlay: novel UX enabling AI to observe screen and audio for contextual assistance.
Cluely vision: generalize the overlay UX beyond cheating to everyday multimodal AI assistance.
Aim: replace prompting with contextual, always-on multimodal AI interaction in 2–5 years.
“Cheat on everything” used intentionally ambiguous to provoke reflection about AI advantages.

Primary constraints: latency (time-to-first-token) and accuracy of model outputs.
Proposed tactics:
- Host models on own servers to reduce external request/load-balancing latency.
- Cache and parameterize inputs to reduce input size and speed responses.
- Craft specific system prompts to improve accuracy.
- Build custom evals from analytics to iterate model behavior.
Personalization plan: gather user-specific data to generate hyper-personalized models tuned to roles and preferences.

Launch timeline: Cluely desktop app launched ~one month prior to talk.
Metrics: nearly $5M ARR and $5.3M seed round closed (lead investors Abstract Ventures, Susa Ventures).
Growth strategy: exploit virality and provocative social content to capture market quickly and secure first-mover advantage.
Moat: combination of unique UX, user personalization data, and speed/accuracy improvements.

LeetCode-style interviews = rote memorization, poor proxy for on-the-job skills.
Technical interviews must evolve as AI can answer many standard questions.
Future hiring may rely on holistic signals or AI-assessed work rather than hour-long interviews.
Universal AI assistance will raise productivity massively, accelerating scientific and societal progress.
Perspective: if AI helps, use it; democratizing AI advantage reduces the notion of “cheating.”

Roy intentionally crafts controversial tweets to drive engagement on X/Twitter.
He separates online persona from private life; close family and trusted friends form his real-world support.
Virality provided social protection and accelerated entrepreneurial path after Columbia incident.
Advice: take progressively larger risks; downside often smaller than perceived, upside larger.

Pivot from Interview Coder (single-use cheating tool) to Cluely (sustainable multimodal AI product).
Prioritize UX-first approach (translucent overlay) as market differentiator.
Invest in infrastructure to reduce latency by self-hosting models when feasible.
Use analytics-driven in-house evals and personalization to build defensibility.

Timeline and plan for migrating from third-party model hosting to self-hosted infrastructure.
Specific privacy and safety measures for an always-listening/screen-seeing overlay.
Regulatory and compliance approach for a product that can be used to “cheat” in many contexts.
Roadmap for personalization: data collection, storage, opt-in, and model fine-tuning processes.