Stripe's internal AI assistant Kai reached 83% weekly employee adoption using LangChain's Deep Agents framework.
“the Deep Agents layer solves all non-Stripey problems, so his team could focus on Stripe's problems.”
context-management
12 tracked signals on context-management.
Screener sub-agents read full traces so the main agent's context stays manageable
“we're really familiar with designing an org chart, in a way”
AI coding agents behave like brilliant engineers with no memory, forcing constant re-explanation of context across repos.
“You would have a genius on one side and something deeply deficient on the other. And that's what agents are.”
LangChain launches Deep Agents, an open-source toolkit for building complex long-running agents
Agent 'skills' are dynamically loaded prompts enabling progressive capability disclosure
“Skills are just prompts, they're fancy prompts, but they're really easy to share and they can help agents really narrow in and do a certain task well.”
OpenAI shares how Jason Liu uses Codex to preserve context across long-running, multi-prompt projects.
“Learn how Jason Liu uses Codex to preserve context, manage complex projects, and help work continue beyond a single prompt.”
Salesforce argues agents need full cross-channel context to answer with correct intent
“Getting that information lets the agent know what's happening. So, when the question comes down, they understand what did the user do previously, what the user is doing right now, so they can answer the question with the right intent.”
Continuous context across agent handoffs separates useful AI tools from throwaway demos
“Maintaining deep, continuous context across an entire handoff is the literal difference between a throwaway demo and a tool that actually changes how you work.”
LangChain defines agent harness as infrastructure enabling perception, action, memory, and skill-loading for AI agents
“An agent is simply a model plus a harness (infrastructure).”
LangChain's managed deep agents support on-demand skill loading via progressive disclosure
“Skills are basically context and scripts that your agent can run and read dynamically as it's running.”
Apple explains agents learn via notes and tools, and Xcode's agent MD file is injected into every query so keep it short.
“You wouldn't want to put like your entire style guide in there every single time. You'll be eating up a lot of tokens.”
Decision records like ADRs help both humans and LLMs overcome limited context and lost institutional knowledge.
“People forget. LLMs context compact. Humans leave. LLMs have no memory.”