LangChain

OpenWiki is a CLI tool that uses a Deep Agents documentation agent to read a codebase or personal knowledge sources and generate a self-updating, linked Markdown wiki. It supports twelve model providers, connectors like Notion, Slack, Gmail, and git repos, CI-based auto-updates, and an interactive node-graph visualizer for exploring the generated wiki.

LangChain engineers detail a Kubernetes SRE agent: scheduled checks skip the agent and use a single Haiku call, read-only subagents fan out for diagnosis, and writes are isolated behind human approval. Traces drove cost cuts and eval datasets. Vendor-authored, with LangSmith framing.

LangChain explains how it built a Slack-based paid media agent on Deep Agents. Covers sandbox workspace design, skills versus wiki context layers, code for computation, per-metric sources of truth, searchable tool catalogs, subagent isolation bugs, and approval-gated actions. Open-sourced.

Agent skills that teach LangChain concepts inside the coding agent you already use.

LangChain's open-source OpenWiki CLI builds a linked Markdown wiki from a repository, tracks facts back to source evidence, and now ships a graph visualizer. Coding agents query it over MCP, including across linked repositories, and updates can run on a schedule in CI.

LangChain opened Managed Deep Agents in public beta. Authors write a Deep Agent in Python or TypeScript and deploy with mda deploy to a LangSmith runtime handling durable threads, sandboxes, memory, Slack channels, identity and Harbor evals. Beta is US-region and CLI-first.

A 17-minute tour of OpenWiki, LangChain's open-source self-maintaining wiki: how it works, why documentation built for agents looks nothing like documentation for humans, and the lessons from building it.

LangChain's walkthrough of regression evals for Managed Deep Agents using Harbor. Each eval pairs an environment, an instruction and a pytest verifier, runs in a fresh container, and logs to LangSmith. Claude Code can scaffold the suite, and it can run nightly in CI.

LangSmith usage data shows open-weight models going mainstream: one in three active teams ran one in July, up from one in five, with daily open-weight usage growing 9x year over year against 4.8x for closed models.

LangChain introduces WikiBench, which scores generated codebase wikis by having a reader agent answer repo-grounded questions against fact rubrics. Wiki plus source beat source alone at lower cost, while wiki alone lagged. Harness cost ranged from $0.44 to $9.18 per run.

Case studies from Lyft, Fastweb + Vodafone, and LATAM Airlines on running customer experience agents in production. Covers trace-driven routing fixes, a supervisor redesign that cut cost about 15%, pass/fail eval rubrics, and calibrating LLM simulated users. Published by LangChain, so tooling is Lan

LangChain says its MCP integration now supports the new stateless MCP spec. The post points to a separate writeup by a LangChain engineer that was not collected, so details on transport behavior and other changes are unconfirmed.

LangSmith's LLM Gateway public beta adds data protection for Enterprise users. It detects PII and secrets such as API keys in model calls, redacts them before they reach providers, restores originals in responses, and scrubs recorded traces.

LangSmith LLM Gateway enters public beta as a governance layer between agents and model providers. It enforces fallbacks across models and hosts, spend caps at four levels, rate limits and PII redaction (Enterprise), with events traced in LangSmith.

OpenWiki 0.4.0 adds coding-agent integrations: init and update can be driven from Claude, Codex or OpenCode with almost no configuration, folding wiki generation into the agent loop instead of a standalone CLI run.

LangChain updates LangSmith Engine, which scans agent traces for issues and proposes fixes. New: self-hosted deployment, Slack and Linear integrations, a Reduced Analysis cost mode, and auto-closing of stale issues. LangChain claims over 2x on its internal IssueBench and 25% on Terminal-Bench fixes.

Similarweb's engineers explain how they evaluate Deep Research reports without golden answers. They combine deterministic tool checks, anchored rubric judges, faithfulness checks and baseline A/B comparisons, and recount a week lost to conflicting rubric criteria.

LangChain's Managed Deep Agents now provision the Slack app as part of deployment. No manifest to author, no OAuth redirect configuration, no bot tokens to copy between environments — one command and the agent shows up in the workspace.

OpenWiki 0.4.0 adds OKF v0.2 support and a claims runtime that keeps a persistent source of truth behind generated pages, surfacing which sources built a page, whether it is fully verified, and what produced it.

Slack's Add to Slack launch makes the messaging app a one-click deploy target for agents built in LangChain, OpenAI, n8n, Lovable, Vercel and other partner tools, skipping custom bot integration work.

LangChain's hosted Deep Agents now build their execution environment at deploy time from a setup.sh or Dockerfile, so each new thread starts on a prepared image instead of repeating clone and install work.

A Unify engineer explains why OpenAI's default prompt cache routing breaks down past roughly 15 requests per second per key, and how custom routing across keys kept cache hit rates near 95%. Based on a short podcast clip; implementation details are not shown.

LangSmith Deployment adds Preview Builds in public beta: each GitHub PR can spin up a short-lived deployment that updates per commit and tears down on merge. Controls include idle TTL, max concurrency, and label-only triggers. Requires GitHub integration on LangSmith Cloud.

Unify's Connor Heggie, speaking on LangChain's Max Agency podcast, says the LLM judge should come from a different model family than the agent it grades. Same-family pairs share distributions and blind spots, so evals drift toward groupthink. This is practitioner advice without measured data.