David Luan
CEO & Co-Founder, Adept AI
Former OpenAI VP of Engineering who left to build AI agents that can use any software tool the way a human would.
Credentials
CEO and co-founder of Adept AI (2022). Former VP of Engineering at OpenAI, where he led engineering teams building GPT-3. Previously at Axon (Taser). BS in Computer Science and Mathematics from MIT.
Why They Matter
Luan is betting on a specific vision of AI that matters enormously for businesses: agents that don't just generate text but actually operate software — clicking buttons, filling forms, navigating apps. If Adept succeeds, it changes the economics of every back-office operation that currently requires a human sitting at a screen. For ASEAN businesses drowning in manual SaaS workflows, the agent paradigm Luan is building could be more disruptive than chatbots ever were.
Positions
AI Timeline View
Believes general-purpose AI agents capable of performing complex multi-step tasks across software tools are achievable within a few years. Focused on near-term practical AI rather than AGI timelines.
Safety Stance
Key Beliefs
The most impactful near-term AI application is agents that can use existing software tools, not replacing software entirely.
Adept AI launch announcement and company vision
AI should augment human workers by handling tedious multi-step workflows, freeing people for higher-level decision-making.
TechCrunch interview
The interface between AI and existing software (action models) is as important as language models — AI needs to act, not just talk.
Adept ACT-1 demo and blog post
Scaling laws apply to action models just as they do to language models — bigger models with more diverse training data will become better agents.
Industry discussions and Adept research direction
Controversial Take
Left OpenAI at a time when it was the undisputed leader in AI research, believing the future belonged to AI agents rather than chatbots. While most of Silicon Valley raced to build ChatGPT competitors, Luan bet on a fundamentally different interaction model — AI that uses your existing tools rather than replacing them. The bet is still unproven at scale.
Track Record
How well have David Luan's predictions held up?
AI agents that can operate software tools would become a major industry focus
Made: 2022
By 2024-2025, nearly every major AI lab (OpenAI, Anthropic, Google) launched agent or tool-use capabilities. The "AI agent" paradigm became one of the hottest areas in AI.
Action Transformer (ACT-1) would demonstrate that AI could learn to use arbitrary software by observing human actions
Made: 2022
The ACT-1 demo was impressive, but shipping reliable agents at production quality proved much harder than the demo suggested. Adept pivoted toward enterprise partnerships.
Key Quotes
“We want to build a general intelligence that can do anything a human can do on a computer.”
“The future of AI isn't just about generating text — it's about taking actions in the real world through software.”
“We're building the model that sits between the user's intent and all the software they use. That middle layer is what's missing.”
“At OpenAI, I saw how powerful large models could be. The question that kept nagging me was: why are we only using them to generate text?”
Publications
ACT-1: Transformer for Actions (Adept AI technical blog)
2022
Last updated: 2026-04-12
←Back to AI Minds Directory