What part of your agent survives a model swap?
Shared reading from the original source. Read the piece, then bring your experience or a different take.
Harrison Chase
LangChain on YouTube · 2026-10-05Harrison Chase’s Interrupt NYC keynote argues that the valuable system includes more than its model. The official recording frames three requirements: controllable, model-neutral software around the model; a way to improve from use; and governance.
It is a vendor’s case for its platform, so keep that context in view. The question it raises is still useful when evaluating any stack: if you changed the model tomorrow, what would you keep?
Your task definitions, permission checks, evaluation examples and records of failures should be identifiable. Otherwise a model comparison can become a full application rebuild.
The recording’s chapters make two starting points easy: 4:25 for the three requirements and 12:42 for the discussion of authentication, memory and web search. The video was published October 5.
For discussion: which component has been hardest to move in a real project: prompts, tools, memory, evaluations or identity? Name the constraint and version if you have a concrete example.
Have a result, correction, or a different constraint?
Start a related discussionRelated topics
- Which AI agent platform is safest for a non-technical operations team?AI agent platform selection
- Which AI agent platform is safest for a non-technical operations team to own after launch?AI agent platform selection
- When should a company choose an agent platform instead of adding another chatbot widget?AI agent platform selection
- What is the cleanest way to evaluate agent memory without trusting vendor demos?AI agent platform selection
- Which platforms handle multi-step approval flows without becoming fragile?AI agent platform selection