Subvolts take Build AI that learns like an employee, not just a bigger chatbot. Jun 24, 2026 - Developers
- Learn about alternative AI architectures focusing on memory and continual learning, potentially reducing token costs.
- Business
- Understand how AI can be trained to deeply understand your company's specific knowledge, acting like a long-term employee.
This interview explores a novel approach to AI memory and learning that could significantly change how businesses use AI, offering a practical alternative to current frontier model development.
Memory and Continual Learning: Engram's Dan Biderman and Jessy Lin
Audio streams from Sequoia Capital. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Engram co-founders Dan Biderman and Jessy Lin discuss their approach to AI memory and continual learning on Sequoia Capital's Training Data podcast. They propose baking company knowledge directly into model weights, rather than relying on large prompts or RAG. This method aims to create personalized, always-learning models that perform comparably to frontier models with significantly fewer tokens.
Imagine an AI that truly remembers and learns about your company, like a long-term employee. Instead of just reading long instructions, this AI gets smarter over time by having that knowledge built in. This could make AI much more useful for specific tasks without needing constant updates.
This episode offers a different perspective on AI development, focusing on personalized memory and learning instead of the race for larger, general models. It's for anyone interested in the practical application of AI in businesses and how models can truly adapt to specific needs. The approach contrasts with frontier labs' current direction, suggesting a future of individual AI assistants that grow with their users, potentially making AI feel more like a capable, evolving team member.
- HostSequoia Capital
- GuestDan Bidermanco-founder, Engram
- GuestJessy Linco-founder, Engram
- Guest and Title: Engram's Dan Biderman and Jessy Lin on Memory and Continual Learning
- Show: Training Data (an official AI show)
- Length: 45 minutes
- Engram's Premise: Bake a team's knowledge directly into the model's weights for memory and continual learning.
- Performance Claim: Matching or beating frontier models while consuming up to 100x fewer tokens.
- Partners: Microsoft, Notion, and Harvey
Dan Biderman and Jessy Lin, co-founders of Engram, are building a neolab around memory and continual learning, which they call two sides of the same coin. Their contrarian premise: instead of stuffing ever-larger prompts into the context window or bolting on RAG, bake a team's knowledge directly into the model's weights, so it knows your company the way an employee of several years does.
The payoff: matching or beating frontier models while consuming up to 100x fewer tokens. Working with partners like Microsoft, Notion, and Harvey, the team draws on roots in computational neuroscience and state-space architectures to attack what they see as the real bottleneck in AI — not raw intelligence, but memory and continual learning. In contrast to the frontier labs' race toward one ever-bigger model and AGI, Dan and Jessy imagine a world where everyone has their own model — privately trained, always learning, and good at the things you actually care about. The real ChatGPT moment for memory, they argue, is the day your model feels like an intern that genuinely got smarter overnight.
Hosted by Sonya Huang and Shaun Maguire, Sequoia Capital
- prompt
- Dan Biderman and Jessy Lin
- the instructions or text given to an AI to get a response.
- RAG
- Dan Biderman and Jessy Lin
- Retrieval Augmented Generation, a technique to give AI access to external information.
- context window
- Dan Biderman and Jessy Lin
- the amount of information an AI can consider at one time.
- state-space architectures
- Dan Biderman and Jessy Lin
- a mathematical framework used in modeling dynamic systems, applied here to AI learning.
1 h 08David Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience
WithDavid Silver, Lasse Espeholt
1 h 041 h 0552 minMaking Cities Awesome: Peregrine’s Nick Noone & Ben Rudolph
WithNick Noone, Ben Rudolph
Questions people ask
- What is Engram's core idea for AI?
- Engram focuses on baking a team's knowledge directly into an AI model's weights, enabling continual learning and deep company understanding, like a long-term employee.
- How does Engram's approach differ from current AI trends?
- Instead of larger prompts or RAG, Engram integrates knowledge into the model itself, aiming for comparable performance with significantly fewer tokens and a more personalized AI experience.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Sequoia Capital.