A startup's primary competitive advantages in the AI era are superior focus on a specific use case and access to proprietary data that is not on the open web.
The most effective way to guide AI model development is through 'evals' (evaluations), which have replaced traditional product specifications as the key tool for defining desired behavior.
Pricing AI products based on usage (tokens, GPU time) is a flawed strategy that leads to commoditization; pricing must be aligned with the value created for the customer.
The future of user interaction with technology will shift away from graphical interfaces and toward proactive, voice-based conversations with AI agents.
Hiring should prioritize a 'growth mindset' and autonomy above all else in the final interview stage, with the expectation that a successful hire will not need to be told what to do after six months.
Early Career (Facebook/Instagram)
Deng's discourse reflects on his time at Facebook, where he was involved in growing Messenger from zero to 4.7 billion daily messages. He also draws lessons from product failures like Instagram's Bolt, using its retention curve as a key metric for product-market fit.
Mid-Career (Uber)
His experience at Uber shaped his core product philosophy, leading him to conclude that the true 'product' for a service like Uber was not the app's UI, but the fundamental value drivers: the price and the ETA.
Recent Past (OpenAI)
As a leader at OpenAI, Deng's focus shifted to the unique challenges of AI product development. His commentary from this period centers on team structure, the role of 'evals' in shaping models, internal debates on UI/UX, and the difficulties of establishing a value-aligned pricing model for ChatGPT Enterprise.
Present (Felicis)
In his current role as an investor, Deng synthesizes his past experiences to form a thesis on what makes AI startups successful. His discourse is now focused on advising early-stage companies on defensibility, market positioning against incumbents, and product strategy.
Future Outlook
Deng frequently makes forward-looking statements, predicting a future where voice interfaces replace many UIs, professional coding becomes an obsolete skill due to AI, and AGI serves as a necessary but insufficient tool for solving major world problems.
▶AI Product Strategy and DevelopmentApr 2026
Deng outlines a modern approach to AI product development where the traditional product spec is replaced by the 'eval' (evaluation), which defines the model's desired behavior. He argues the highest leverage for product managers at foundation model companies is working directly with research teams on post-training, rather than on the application layer.
This signals a fundamental shift in the product manager role in the AI era, requiring deeper technical understanding of model behavior and evaluation over traditional UX/UI specification.
▶Startup vs. Incumbent Dynamics in AIApr 2026
Deng believes startups have a significant advantage over large model companies by focusing entirely on a specific use case, allowing for greater speed and product craft. He identifies three key defensibility moats for AI startups: unique proprietary data, a data flywheel from user interaction, and rapid market penetration.
For investors, Deng's framework suggests that a startup's access to non-public data and its ability to execute on a specific workflow are more critical long-term indicators of success than its initial choice of foundation model.
▶The Business and Pricing of AIApr 2026
Deng is highly critical of usage-based AI pricing models (e.g., per-token or GPU consumption), believing they create a race-to-the-bottom commodity market and fail to align with customer value. He found per-seat enterprise models flawed for the same reason and predicts that subscription models will need to evolve dramatically.
This perspective indicates a significant unmet need in the market for innovative, value-aligned pricing strategies for AI products, presenting an opportunity for companies that can solve this challenge.
▶Philosophy on Talent and HiringApr 2026
Deng's hiring philosophy centers on finding individuals with a 'growth mindset' and a high degree of autonomy. He uses a six-month rule: if he is still telling a new hire what to do after that period, he considers it a hiring mistake, entrusting his team to vet for other skills.
This approach prioritizes long-term adaptability and self-direction over immediate, specific skill sets, suggesting that in a rapidly changing field like AI, the ability to learn and operate independently is the most valuable trait.