Engram emerges from stealth with continual learning AI that cuts inference costs by training models to know your world

Jack Morris is co-founder and head of research at Engram, which emerged from eight months of stealth last week with a continual learning AI system and backing from an undisclosed group of venture investors.

The core idea is that frontier labs compete to make one model smarter every month, but Morris argues that's the wrong axis for most enterprise use cases. What enterprises actually need is a model that knows their world better. Engram's system continuously retrains itself on a customer's specific context, effectively rewiring the model's knowledge daily rather than relying on generic capabilities.

“The model doesn't need to get smarter every month — it needs to know you better. We're working on a whole different stack: new ways of training, new ways of running the models that train themselves to know your world better and adjust to the things that you say. Our early enterprise partners that we've been working with are Microsoft, Notion, and Harvey. They're nice because they have these massive workspaces of context and they're early adopters of AI — these are the places where we can reduce costs the fastest because the workflows really are just that repetitive.”

The cost argument

The near-term commercial thesis isn't about smarter AI — it's about cheaper AI. Enterprise agents running repetitive workflows typically re-read large batches of context files from scratch on every run. Morris says Engram's approach cuts that dramatically: where an agent might read 100 files to generate a daily summary, a continually trained model might need only four. That reduction in inference cost is how Engram is finding early traction.

Design partners named at launch are Microsoft, Notion, and Harvey. Morris singles them out precisely because their workflows are highly repetitive and their workspaces context-heavy, which is where the cost savings compound fastest.

The longer bet

Morris frames the inference cost play as a near-term wedge. The longer ambition is a model that develops a genuinely deep understanding of a user or organization over time — the way a long-tenured colleague would — and can generalize across tasks rather than just retrieve known context faster. Whether Engram can sustain a technical edge as the frontier labs iterate on their own context-handling is an open question the transcript doesn't address.