Loading activity…
@sydneyrunkle
product + open source @LangChain
AI
A post describes Harness² refining prompts, tools and memory, and reports an 11.5-point WorkBuddy-Bench gain with frozen model weights.
AI
A developer says the schemas validate human responses before they return to the agent and can help render approval forms.
AI
A post announcing the integration lists model-routing and auto-mode middleware among the features shipped.
AI
LangChain says agents can find products, prepare purchases and pay, with a person approving every order.
AI
LangChain says bound tools stay hidden until an agent reads their skill, keeping unused tools out of context.
AI
LangChain lists four steps, from understanding tasks and models to tracking task outcomes.
AI
A post announcing the update lists DOCX, PPTX, XLSX, CSV, JSON and Markdown support on OpenAI Responses, beyond PDFs.
AI
A post says the feature lets agents remember the people they work with.
AI
A post argues that every agent harness needs a durable runtime, pairing pi with pi-durable and deepagents with LangGraph.
AI
LangChain says the router chooses among three model tiers using each thread’s first message, rather than defaulting to its strongest model.
AI
A user says human touches per pull request are easy to track, while scoring a PR is difficult.
AI
A user says they start the agent in Slack, then their GitHub review comments automatically trigger it to address the feedback.
AI
A post says developers can define valid responses when an agent pauses for human input. Responses are validated before they’re passed back to the agent.
AI
LangChain says agents can take requests from any webhook, not just Slack. It highlights a PhotonHQ-backed expense agent that logs receipts sent by text message.
AI
A participant says the conversation covered how to use the new MCP spec with FastMCP and LangChain, why evaluations are hard, and decision models like jev.
AI
LangChain reposted a user's recommendation of LangChain and LangSmith for people building AI agents.
AI
LangChain says Jev was 5–6x faster than Sonnet at classifying pages in its document-review demo, with LangGraph routing pages to an LLM or an attorney when needed.
AI
A user says SemIf is much faster so far than their previous LLM classifier for labeling incoming repository issues. They're still monitoring whether the threshold for applying labels is well calibrated.
AI
A user shares a guide they describe as explaining what makes jev special and how to use it with agents.
AI
OpenRouter says Typesafe's Jev returns structured decisions with probabilities rather than generated text. A browser-agent creator reports using it with Browser Use to find flights in 7 seconds for $0.0039.
