Knowledge catalogue
This post emphasizes a three-stage lifecycle for AI/LLM applications: evaluating systems before deployment to ensure quality, monitoring their performance in production to catch issues, and using the
Gall’s law: “a complex system that works is invariably found to have evolved from a simple system that worked.” also very relevant to agent systems. most teams are trying to jump straight to autonomou
Harrison Chase suggests that AI agents designed to handle knowledge work tasks will probably require browser capabilities to effectively access and interact with web-based information and resources. T
@hwchase17 the tooling side is maturing fast but the feedback loop between engineers and domain experts is still the bottleneck. keep running into this building agents: you can instrument everything a
my fave point from here: the earlier you think about your agent as a system that can be measured & improved, the faster you can get a robust agent into production This isn’t just a technical thing, it
one way to view langsmith is as a platform for the whole org to collaborate on building agents helps speed up that feedback loop between different personas @hwchase17 the tooling side is maturing fast
Spot on regarding virtual filesystems. Too many agent demos default to full sandboxes or local environments. When you actually deploy a service, data isolation, performance, and stability become bottl
This post from Harrison Chase discusses how AI agents have evolved beyond just the underlying model, emphasizing critical infrastructure components including state management, observability, retry mec
Top AI development teams prioritize rapid iteration and early deployment of agents rather than pursuing perfection before launch, as continuous improvement based on real-world feedback is critical to
viv always says it better than me lots of talk recently of thinking of agents as systems to measure and iteratively improve but - thats not JUST a technical thing. its also a human & team thing my fav
ai coding is getting expensive use more open models! Kimi K2.6 on @baseten is ~5x cheaper than Opus 4.7 For a large majority of tasks, it's roughly the same performance If you want to use open models
.@BraceSproul changed our org's internal model in Fleet from Sonnet 4.6 to Kimi K2.6 and I didn't even notice. Open models are already good enough for most tasks, though not the hardest coding work ye
concluding my 'taking deep agents to production' series with arguably the most important component: observability. when you deploy a deep agent with LangSmith, you automatically get traces for every r
🔒🔓Configurable tracing in Fleet agents You can now enable or disable tracing on a per-agent level in Fleet! This is a big unlock for developers who can now get all the trace details for their Fleet ag
I feel like we're all circling around the right unified expression of self-optimizing memory+skills+evals and in the next 6 months or so someone will nail it and unlock the next big step change in age
i keep coming back to this post -- it enumerates a ton of requirements you need to think about when taking an agent to production, and how the langgraph runtime is built to address these needs! https:
Kimi K2.6, deployed on Baseten, offers approximately 5x cost savings compared to Claude Opus 4.7 while maintaining comparable performance on most tasks, making it a viable open-source alternative for
Open-source language models have reached sufficient quality and capability levels just as commercial frontier model providers have begun increasing their inference API pricing. This timing creates a p
this is the part people underestimate about agents the “prompt + tools” part is honestly the easy bit production agents need: - durable execution - memory - human-in-the-loop - retries - concurrency h
'Tis the year of open source LLMs in agents! .@BraceSproul changed our org's internal model in Fleet from Sonnet 4.6 to Kimi K2.6 and I didn't even notice. Open models are already good enough for most
LangChain's Interrupt event in San Francisco is expecting approximately 1,000 attendees across Wednesday and Thursday, with organizers anticipating tickets will sell out by the end of the week. The ev
I cannot complete this request because the URL provided appears to be invalid or fabricated (the status ID format seems implausible), and the tweet text is incomplete, cutting off mid-sentence. Withou
BYO sandbox with deepagents! this is the part of the deep agents production series i've been most excited to get to: sandboxes without an execution environment, a production agent is only as capable a
Episode 3 of Max Agency is here! Join @hwchase17 + @Ramp’s Head of Applied Research Alex Shevchenko and go behind the scenes of Ramp Sheets. @RampLabs Talked to @ramplabs Head of Applied Research Alex
Most teams can build agents, but far fewer have the infrastructure or know-how to run them reliably in production. Join us for an evening in San Francisco on May 19th, with Victor Moreira + @Vtrivedy1
my posture is terrible but Alex's insights are great, so it evens out Talked to @ramplabs Head of Applied Research Alex Shevchenko on the Max Agency podcast to learn how @Ramp Sheets was built, their
New Max Agency with @tryramp Talked to @ramplabs Head of Applied Research Alex Shevchenko on the Max Agency podcast to learn how @Ramp Sheets was built, their internal agent Inspect, and so much more.
one of the features i'm most excited about in our upcoming langgraph release is delta channels! the langgraph runtime lets you 'checkpoint' agent progress at every step (model call, tool call, hooks).
Open Models Make Agentic Batch Processing Economically Viable A lot of world’s work looks like “Do X for EVERY Y” - read every trace - respond to every email - deep dive into every document - enrich e
Open Models x Headless Agent Execution 🔥 your daily reminder that open models are plenty capable for a lot of coding work. easiest place to feel that out is deepagents! swap the model and go. i've bee
Ramp Labs is doing a ton of cool work all the way from launching products to hardcore research on LLM internals, this is a fun one 🔥 Talked to @ramplabs Head of Applied Research Alex Shevchenko on the
Talked to @ramplabs Head of Applied Research Alex Shevchenko on the Max Agency podcast to learn how @Ramp Sheets was built, their internal agent Inspect, and so much more. YouTube: https://www.youtube
this is the part of the deep agents production series i've been most excited to get to: sandboxes without an execution environment, a production agent is only as capable as its fixed toolset. give an
This appears to be a promotional or event announcement post shared by Harrison Chase on X (formerly Twitter) linking to a Luma event page, likely advertising an event happening in New York City. Witho
300 million agents. @clay runs all of them through LangSmith. This is what production-grade AI looks like at one of the fastest moving startups! .@Clay uses LangSmith to manage 300M agent runs a month
A great conversation between Noah Kravitz from the @nvidia team + @hwchase17. “Every enterprise needs a claw strategy.” How did @LangChain go from a weekend project to 1B+ downloads in 3 years? We sat
A TLDR on Harness Profiles: ✅ Model-specific profiles to adjust prompts, tools, and middleware. 📦 Profiles for @OpenAI, @Anthropic, and @Google models out of the box. 📈 A 10–20 point jump on a subset
Anthropic is building out their managed agents platform, adding Dreaming (memory) and Outcomes (rubrics). The idea I'm wrestling with: how differentiated are these platform features really? I initiall
.@Clay uses LangSmith to manage 300M agent runs a month, with an average 10-30 steps each. @hwchase17’s conversation with Clay’s Head of AI @jeffbarg on how they run this at scale → https://youtu.be/c
“Every enterprise needs a claw strategy.” How did @LangChain go from a weekend project to 1B+ downloads in 3 years? We sat down with CEO and co-founder Harrison Chase (@hwchase17) to talk deep agents,
It feels like agent harness evolution runs on two axes that usually get conflated. There’s the temporal axis: simplify as models improve, stripping components that compensated for limitations the new
.@ListenLabs Co-Founder & CTO @florian_jue on the 2 types of engineers who build great agents: 1/ Those who know what LLMs can and can't do, and can feel when something's off 2/ Product engineers who
really cool work by the Harvey team, excited to partner with them to push forward research on designing + understanding agents across Long Horizon Legal work the first peak I got at LAB was 'woah this
small workflow note that adds up. /staged-pr is a skill (via slash command) i run when wrapping up a PR. it takes my staged code changes and drafts a concise PR title & description, based on my prefer
Strong Opinions, Loosely Held on Agent + Harness Engineering: 1. You can outperform any default harness+model (including codex & claude code) on pretty much any Task by engineering the harness around
the deeper point: model choice and harness choice aren't independent variables. benchmarking a model without specifying the harness is like benchmarking a chip without specifying the compiler. the num
this deepagents deploy https://docs.langchain.com/oss/python/deepagents/deploy (or at least directionally where we want to take it) what's missing? give us feedback! can someone PLEASE launch OS claud
Harrison Chase expresses enthusiasm about collaborating with Harvey on their Long Horizon Legal Agent Benchmark, highlighting the value of developing industry-specific benchmarks for AI evaluation. Th
Harrison Chase expresses support for expanding AI benchmarking efforts and highlights appreciation for work being done by Harvey (likely Harvey AI, a legal AI company). The post indicates enthusiasm f
we're continuing to see clear examples where a model's harness is a major determinant of overall performance. with the same model, running on same task, it's easy to observe very different scores depe
A few days ago i wrote a post about how we use the LangGraph checkpointer to optimize storage. Today, LangGraph released v1.2 with a really nice feature: DeltaChannel, a new channel type that stores o
Agent observability enables monitoring of agent behavior, but to effectively improve agents through feedback loops, developers must actively collect observability data and generate synthetic examples
Come and join us, we're hiring across almost all roles 🚀https://www.langchain.com/careers Companies I'd consider going to if I ever had the urge to try something new. 1 Thinking Machines Lab 2 OpenAI
everyone would have a deeper appreciation for Agent Products that rock because of great Context/Harness Engineering if they… talked to: - LLM Base models - Post-Trained models with no harness (no tool
Great new drop by @LangChain’s @hwchase17 “If you do not know what the agent saw, what it did, and what happened next, you cannot reliably know what to improve” Traces give agents this insight and tha
I detected a bad Agent action, what do I do about it? this is pretty much the main question that will power the future’s Human+Agent driven improvement loops Gather data -> Mine Errors -> Find out whi
if you haven't read this one by @Vtrivedy10, it's a must read! great overview of what components a harness needs to support an agent for long running, long context tasks https://www.langchain.com/blog
Large language models are becoming increasingly costly to develop and operate, creating a need for open-source alternatives that reduce dependency on expensive proprietary models and make AI more acce
New from the LangSmith Signal: The vector space rapidly diversified since December. ✅ Managed cloud surged. MongoDB Atlas, Pinecone, + Qdrant tripled their combined share to 21%. ✅ Only 3 vector store
Not quietly. Loudly! 📣📣📣 deepagents-cli🔥 deepagents-cli is quietly becoming the best place to start coding with open weight models. we've been investing heavily in making it a harness that's truly mod