开启时会模糊预览图，关闭后正常显示
this concept of engineering harnesses to “shape jagged intelligence” is one my favorite mental models for extracting intelligence from today’s systems every phase from pre-training to RL to harness design is an exercise in shaping the behavior of the shoggoth of these, the most accessible shaping behavior is in agent/harness design - things like prompt design, context engineering, tool design are levers that amplify certain dimensions from the training settings (ex: match the formatting or problem formulation seen in post-training) or shackle the model to so it operates in high intelligence where its intelligence (ex: low context use) we’re trying to set up this incredibly intelligent system for success by designing its world around it, interesting how we (try to)do the same to help people succeed in the real world
leading deepagents and building evals @LangChainAI, prev @awscloud, phd cs @ temple
it’s kinda hilarious that the infra pattern for agents is durable execution and serverless sandboxes
leading deepagents and building evals @LangChainAI, prev @awscloud, phd cs @ temple
Skills and Continual Learning are in rn 👀 @RLanceMartin does an awesome job going over how agents can learn through experience covering patterns we find useful with DeepAgents (and agents generally) including - turning sessions into Skills that can be shared across teams - using agents to mine sessions to improve prompts/instructions - updating and consolidating user memories the loop of Build Agent —> Run Agent —> Analyze Agent Outputs —> Edit Agent is a rlly powerful agent engineering improvement flywheel
leading deepagents and building evals @LangChainAI, prev @awscloud, phd cs @ temple
if 2026 has more ppl doing Karpathy style single file, low to no abstraction software it’ll be a great year :) not always the way but great way to start
leading deepagents and building evals @LangChainAI, prev @awscloud, phd cs @ temple
everyone in ai is cooking some holiday side quests/projects gonna be some fire demos/products in Jan 👀
leading deepagents and building evals @LangChainAI, prev @awscloud, phd cs @ temple
it’s kinda obvious useful to think about for agent + harness —> progressive disclosure for skills disclosure is a feature of the harness, not the model the model didn’t choose to have all of the YAML frontmatter pre-loaded on its first turn the nice folks at Anthropic came up deterministically designed this pattern to help the model be better with context this mode of thinking is the recipe for good harness design basically what can we include something for the model (might be a subagent, a context management pattern, infra, etc) so that it can do better across the tasks we care about
leading deepagents and building evals @LangChainAI, prev @awscloud, phd cs @ temple