Harrison Chase

RT Ajitesh Shukla Building an AI agent? Write the eval runner before the agent. We're building an agent using @langchain deep agents that answers vendor invoice queries in procure-to-pay (P2P) workflow, and the eval runner is the first thing we're writing. Why it comes first: • You have to define "correct" before you write a single skill • Every skill, model or tool change is re-scored on the same golden dataset • You improve the agent against numbers, not gut feel So the agent depends on Ports, small Python interfaces, and we plug in mock adapters that return fixed test data: class InvoicePort(Protocol): def get_status(self, invoice_id: str) -> InvoiceStatus: ... class MockInvoiceAdapter: def get_status(self, invoice_id): return FIXTURES[invoice_id] The agent can't tell it's talking to a mock. What you get: ✅ Repeatable evals that give the same result every run ✅ Agent work starts on day 1, not when access is granted ✅ Real adapters replace the mocks later, with no change to the agent or the evals ✅ The frozen mocks become your regression suite Evals first. Ports second. Agent third. #AIAgents #EvalDrivenDevelopment #Python #AgenticAI #LLMOps

RT Viv a view into the future of how humans will design work & allocate compute to make it happen if we make it easy for agents to achieve goals via self-verification & good tooling…a lot of cool stuff can just exist

Come run with us

RT Mihir Garimella Today, most agents - even proactive ones - are defined by the specific task you want them to take on. In the future, agents will work alongside humans to own longer-term goals (and do a wider range of work to get there). The hardest part is shaping these goals so the agent can pick up the context it needs - especially subjective things like taste and judgment that can only be learned from feedback. @hwchase17 & I wrote about lessons learned building these agents. Goal engineering is the new context engineering!

RT Anshul Gupta Prompt -> Context -> Goal. The next evolution in connecting agents to real outcomes!

RT Josh Rosen I endorse the idea of goal engineering for agents with one caveat: Rather than one goal, an agent likely has a collection of more granular goals it must hit along the way to achieving the final goal. So goal engineering is really about aligning a series of goals and determining the dependencies between those goals.

the key insight i learned from @mgarimella was to design goals around what you can give feedback on more easily feedback (used to align the agent with your mental model of what it should do) is the hardest thing to extract from users so the more you can design your system (ui, goals, etc) around making that easy to do - the better!

as agents move towards working on goals, it becomes important to design those goals correctly wrote with @mgarimella on some lessons learned at actively! https://x.com/i/article/2121382906820243456

RT Sai Pittu Own your intelligence! Very much aligned with LangChain's vision here. The open frontier ecosystem has reached the point where you can compete with frontier labs in your own domain. Best example is how Harvey post-trained an open model with Applied Compute for their Review Table product with their own evals and scored above Fable 5, at less than half Sonnet 5's cost per cell. You don't need a lab. You need your domain's data, your evals, your custom harness and an open model. Best time ever to build your own intelligence layers.

RT Ajitesh Shukla Welcome Langsmith platform! Exciting times ahead for @LangChain builders, for one and all! My key takeaways for Langsmith platform from what @hwchase17 spoke in this video:. 1. Agent Development: Teams can build agents using natural language or code and deploy them with a single click. 2. Traceability and Evaluation: Users can monitor every step an agent takes and use automated evaluations to test changes before deployment. 3. 24/7 Monitoring via LangSmith Engine: This feature provides 24/7 monitoring to detect failures and automatically generates fixes and tests for review. 4. Governance: The platform includes built-in controls such as model fallbacks, cost management, and protection for sensitive data. #AgenticAI #AIAgents https://youtu.be/NeV6MSN6zEE

we're celebrating this fundraise with a ton of product launches next week, and then a webinar on friday to wrap it all up come join us! https://events.langchain.com/webinar/future-of-agents/#Registration

RT Sonath LangChain raised $300M at a $3B valuation. The pitch is LangSmith, a platform to run, improve and govern agents.

RT Nick Hollon big news!! we just raised our series c round to help every company own their intelligence. i joined to push what agents can do and get them into more people’s hands. excited to be part of a team that’s all marching toward the same goal at light speed!!

RT Hari Dhanushkodi I joined @LangChain in March 2025, when we were around 40 people. A month later, we shipped Agent Server (then LangGraph Platform). Since then, it’s run 100M+ customer agent executions, and it powers all of our internal agents too! Grateful to build this with an incredible team. Excited for what’s next 🚀

RT Mason Daugherty One of the best teams ❤️🐐

RT Clay Berger When we first invested in @LangChain, they were just starting to convert their massive OSS community into a real business. Since then, they've completely run away with the agent observability market and evolved into the most comprehensive agent engineering platform across runtime, improvement, and governance. We at @IVP are thrilled to be doubling down in the Series C!

RT Neil Dahlke When I joined LangChain in 2025, Deployed Engineering was a handful of people and customers asked if agents could do the thing. We’re past 50 DEs now, and they ask how to run lots of agents and prove with evals they still work. $300M Series C today.

working with jess is very fun and you should want to do so as well! come join us!

RT Jess Ou It’s been a a wild, thrilling, and fulfilling journey from joining when we had just a couple hundred thousand in ARR and less than 20 team members to now over 400, and over 7,000 LangSmith customers. come join us at @LangChain: https://www.langchain.com/careers

RT @ghumare64: Huge congrats 👏

RT Julia Schottenstein Today, we’re announcing @LangChain's Series C fundraise at a $ 3BN valuation, and holy f we’re doing it!! When I joined @hwchase17, @ankush_gola11, and the rest of the 8 person team in the summer of 2023, we were pre-revenue, pre-commercial product, and honestly the biggest risk I saw to joining the insanely stacked crew was “would this LLM technology thing ever do anything useful?!”. I was an alien to Harrison. He didn’t know what to do with me since I wasn’t going to code and that was pretty much all the team did back then. We were so locked in, the small office was silent while hundreds of PRs were handwritten (no coding agents back then!). Later, I found out people referred to our first office as the SAT testing center because yeah… locked in :). Since product was Harrison’s, I did sales and the rest. And day after day, the LangChain team just kept at it. Startups and LangChain are built on belief. I believed this small team could build something great. We believed that more and more software would look like agents, and we were going to power it. Our earliest customers – Carlos at Cisco, Joel at Workday, Yusuke at Rakuten, Tanvi at LinkedIn, Yaron at Klarna believed that betting core parts of their company AI strategy on the Lang stack would work out. Our teammates (many of my friends) who joined us, believed they could grow personally and professionally alongside the company, and our investors @chetanp, @sonyatweetybird, @tomloverro, Mo, Angela, @rajeevdham, @milesgrimshaw continue to believe that we’re still just getting started, and there’s so much growth ahead. We’re a lot bigger and louder than those early days (400+ teammates, 7,000+ customers, products that actually work). But job’s not finished, and I believe / hope it never will be.

RT Sofia Montoya As a former VC, it’s easy to glance past headlines like this. Being on the inside has taught me that momentum is never a given, growth can be painful, and talented people show up every day to make milestones like this a reality. Incredibly bullish on the future - come join us!

RT Jake Broekhuizen Huge milestone for LangChain! Been lucky enough to see firsthand how teams like Harvey, Listen Labs, NVIDIA, Lyft & more are using LangSmith to build better agents. Super excited about what we're building with SmithDB and Engine to help teams continuously improve their harnesses, and even more bullish on using those same traces to help teams post-train & own their own intelligence. This fundraise is a massive vote of confidence in what's ahead. Couldn't be prouder of this team! We're hiring across almost all roles, come join us!

RT LangChain LangChain has raised $300M at a $3B valuation to advance our mission to enable every company to own its intelligence. https://www.langchain.com/blog/series-c More than 7,000 companies already rely on LangSmith for agent runtime, improvement, and governance. We're excited to make it easier for every company to own its intelligence.

RT Viv if any friends have 3-5 mins would love to hear how you’re using & building agents in the wild today and what’s easy/hard/boring/fun/terrible/exciting about it tons ideas & excitement come from us talking with customers, builders, cool ppl doing cool things with agents a lot changed this year, would love to hear how we can better help support y’all ❤️

Persistent agents is a great noun

RT Sherwood Today we're launching Sazabi Security 🔒 Sazabi Security watches your logs for attacks, investigates anything sketchy, and pings you when something is wrong.

RT Ajitesh Shukla In this keynote from Interrupt London, @hwchase17 discusses the evolution of AI agents and the necessity for companies to own their intelligence by building specialized systems around LLMs I loved everything told around deep agents and evals. And, yes, thankfully, we are also talking about decision models such as JEV. Here are my takeaways: 1. The Three Pillars of Ownership: To effectively "own your intelligence," agents must be open, hill climbing and governed. 2. Managed Deep Agents: Combines business logic, the agent harness, and infrastructure (like sandboxes and secret management) into a single, scalable deployment platform. 3. Observability & Evals: LangSmith treats agent traces as a unique data type. It features SmithDB for specialized storage, Jev for efficient decision-model evaluation, and new AI Search capabilities to identify logic failures in trace data. 4. LangSmith Engine: An automated "proactive agent engineer" that monitors traces, clusters issues, proposes fixes, and uses red teaming to stress-test agents. It has already scanned over 70 million traces and detected over 21,000 issues. #agenticai #AIAgents https://youtu.be/XOUSuxY074U

RT Ajitesh Shukla I will be using @langchain deep agents for one of the procure-to-pay POC I am working on. Here is why I would recommend Deep Agents. It bundles the harness with necessary production infrastructure, such as: 1. Capabilities such as designing agents with reusable skills 2. An open-source, model-agnostic harness used for context management and loop orchestration 3. Context Hub: A UI-driven storage solution for instructions and memory allowing to update agent context without redeploying. 4. Harbor: An industry-standard, open-source format used for agent evaluation. #agenticai #AIAgents

c̵o̵n̵t̵e̵x̵t̵ ̵e̵n̵g̵i̵n̵e̵e̵r̵i̵n̵g̵ ̵h̵a̵r̵n̵e̵s̵s̵ ̵e̵n̵g̵i̵n̵e̵e̵r̵i̵n̵g̵ goal engineering

we're running our annual "State of Agent Engineering" survey! we're all just trying to figure out how to build agents together - these are a few of the questions/topics that are top of mind of us excited to see what the community thinks! https://form.typeform.com/to/U2EAO1Lz?typeform-source=t.co

RT Caspar We're running a MDA workshop in SF next Tuesday! Bring a laptop and leave with a deployed agent https://langch.in/mda-workshop

RT Ajitesh Shukla Eval-driven development (EDD) is TDD for AI agents. Great article from @LangChain Stop tweaking prompts and guessing whether things got better. Measure every change against a fixed set of evals. Here’s how to do it 🧵Read the thread to learn. #agenticai #AIAgents https://www.langchain.com/blog/what-is-eval-driven-development

RT Ajitesh Shukla In this session from Interrupt London, @hwchase17 discusses the "North Star" vision of building AI agents that improve over time. He defines an agent as a system consisting of three core parts: 1. The Model (the reasoning engine), 2. The Harness (the surrounding control system) 3. The Context (memory and knowledge). Key dimensions of agent improvement: * Model: Improved by updating weights through Supervised Fine-Tuning (SFT) for simpler tasks or Reinforcement Learning (RL) for complex tasks. * Harness: Involves refining the code, prompts, and tools. LangSmith Engine helps automate this by monitoring traces to identify issues and propose fixes. * Context: The easiest dimension to update; involves evolving memory at the user or organizational level. Foundations for improvement: * Traces and Evals: Harrison argues these are the critical foundation for all improvement. He highlights the need for specialized tools like SmithDB and LangSmith AI Search to handle the unique nature of agent trace data. * Evals as the new PRD: Evals are essential for defining "good" behavior, though they are difficult to build. He covers the challenges of creating realistic inputs, defining correctness, and running simulations without impacting production systems. * Red Teaming: Proactively identifies agent mistakes before they reach production, using algorithmic probing rather than relying solely on user feedback. The Future: * The ultimate vision is a "flywheel" where an agent can notice its own mistakes, create its own evaluation data, and suggest its own fixes to its model, harness, or context. #AgenticAI #AIAgents How to Build Agents That Improve Over Time: Model, Harness, Context https://youtu.be/xy_7SzwsIjM

RT Simon Budziak Everyone is converging on the same agent architecture. An LLM in a loop, calling tools, inside a sandbox. That does not mean the hard problems are solved. @hwchase17 at @LangChain lists what is still open for always-on agents. A proactive agent has to help without annoying your team. Too many notifications and people mute it. Too few and nobody feels the help. Memory has the same trap. Remember nothing, or remember every tiny detail, and the agent gets worse either way. Which of these would hurt most in your company?

Long tasks shouldn't block the chat https://www.langchain.com/blog/agents-are-going-async https://x.com/LangChain_OSS/status/2119173281018171741

RT Caspar Managed Deep Agents can now be steered mid-turn. If you send a message while the agent is working, it reads it at its next model call instead of waiting for the turn to finish. Small change that makes agents much, much nicer to work with

most of the agents we're building inside langchain leverage async subagents in a meaningful way now available in deep agents!

RT Sydney Runkle as we ask agents to do more work for longer, async subagents become a very important primitive async subagents enable - continuous human interaction w/ the main orchestrator, since it's not blocked by subagent work - efficient distribution of work across the background workers

RT Sydney Runkle https://x.com/i/article/2118433978491269123

RT Mike Johnson Re @hwchase17⁠ - we’re building the infrastructure for workflows where AI agents can’t afford to get things wrong, starting with hiring and compliance. We’re hiring to help build it. Thought it might resonate with some of the builders in your community. Would <3 a share if so.

RT Sydney Runkle if you give your agent a browser, it can do work on the web, even on sites w/ no agent native APIs! in this demo my agent books dinner reservations for me. i only had to log into the reservation site once; it saved my login and acts on my behalf for every new session!

RT Caspar A year ago I wrote that LangGraph isn't going anywhere. Today it's the runtime under everything at LangChain, including MDA @JoshARosen wrote a good piece on why deterministic layers will become standard agent infrastructure. We made the same bet early: https://x.com/JoshARosen/status/2117687393213932003

RT Sydney Runkle we're hosting an event on thursday at the @LangChain office with @allietheicon from @typesafeai to chat about all things jev! come learn about - where jev fits in the agent loop - why jev is a "system one" model - jev as a judge for evals https://luma.com/iuc3ostf

earlier this ~spring, models got good enough to drive browsers in some ways, browsers are the new terminal! a general purpose tool one difference is there is a few different ways to design tools to drive the browser. @sydneyrunkle deep dives into how we did this for MDA

RT Anish Agarwal We just raised $120M, with a $1B valuation, to build software that keeps itself running. AI is writing more code than ever. It’s also making production less reliable: more code, shipped faster, understood by few. @traversal_ai understands how production systems fit together, finds the root cause of incidents, and helps prevent the next one. Engineers at Nubank, Harvey, PepsiCo, DigitalOcean, and a growing list of Fortune 100 companies use Traversal every day. Here’s production without the drama 👇

RT Caspar You should build your company harness on MDA https://langch.in/mda

is org/agent level memory just "skills"? assuming there is process for updating skills from agent traces/history why or why not?

RT Git Maxd If you get to go to one conference in 2027 about Applied AI - this is the one You’ll learn from and get to talk to @LangChain Engineers and Ambassadors who provide direct, actionable answers, not a “let’s connect”! Join any of the included cutting edge, hands-on interactive demos and learn directly from the team who applies AI in real-business - it’s the strongest learning op in the next 6 months 🦜🔥. You could be here!

RT Caroline di Vittorio This is a huge reason I believe in cloud agents over local agents. A local agent only runs when you start and watch it, but what will make agents more useful next is proactive, capable work that happens in the background.

We're bringing back Interrupt for the THIRD year - This will be the biggest one yet (new venue that can hold ~2k people!) - Most talks & workshops we've ever had - New Executive Track Register today for the early bird special: https://interrupt.langchain.com/

RT Sydney Runkle we think a lot about context engineering at langchain there's a lot of of old but good primitives here: skills, tool search, summarization to name a few this short blog covers new challenges in this space: browser use, long term goal tracking, async work, and more!

RT Logan I love that agent engineering keeps getting new weird problems as these things evolve

RT Viv personally have the exact opposite view here! building a great harness is a deeply core part of building a great product countless examples you can ask an agent to explain from Grok Bot YouTube video, Factory blogs, our blogs at LangChain, OpenAI’s Codex blogs, etc a great harness is a reflection of building a great experience for the user interacting with your agent. it includes - adjusting prompts (which are off part of the harness) - creating bespoke tools/skills that make common workflows a single tool call -> save on context!!! - aggressively examining thresholds for compaction for your given use case - checking if some work can be offloaded to cheaper models in the harness via subagents/routing - building connectors so agents can efficiently access and search over useful context - supporting background work so users get immediate feedback while other work continues to chug along - building memory in a way that matters for your users ^ this is a small portion of the entire world you think about when building a harness for your product Harness Engjneering helps your team build a better product and better understand of what makes your model tick for your use case also every team building frontier vertical products is harness engineering -> it’s absolutely an edge and it’s fine if the new model makes part of the harness unnecessary layer -> there’s so much optimization you can do at every scale across cost, latency, intelligence related note: the importance of having good evals is the ability to quickly hill climb your harness with new models and trust that the config ex: auto research your harness on a good eval set you trust the frontier models can do a lot -> a lot of tasks can just be one-shotted, but many can’t at the intelligence-cost-speed Pareto you need

RT Maxime Gréau 💯 A major shift toward proactive, always-on, goal-oriented agents is starting now and will gain momentum over the next few months

want to learn how to use decision models with langgraph/deepagents? we're doing a meetup with Typesafe (Jev creators) next week come join! https://luma.com/iuc3ostf

A lot of content engineering challenges have pretty standard solutions, which has lead to some engineers saying agent engineering is solved/boring True to some extent, but there are a bunch of (fun) new ones emerging as agents evolve! https://x.com/i/article/2116547440857718784

when people think of "agent observability" they may think of "datadog for agents" it's really more like "databricks for agents" we've done some stuff in this direction but expect to see more soon

if you can define what good looks like for an agent for a particular domain, you will best positioned to win that domain thats why im so bullish on benchling for bio

RT Nicholas Larus-Stone I hate to rain on everyone's parade, but unfortunately LLMs are still not that good at biology :(. Even basic molecular cloning tasks are beyond the reach of frontier models

a lot of the challenges in harness engineering are now best solved by closer integration of harness and infra async subagents and browser use are two examples of this now in MDA 0.10!

RT nathan drezner most of the web has no API, just pages built for people clicking through them so! we added a browser to Managed Deep Agents. and when a site wants a login, agents can hand over the window so passwords stay out of the model. all built in with every MDA https://x.com/i/article/2115851731741347840

RT amal a lot of you asked how i made our launch video for Content Studio it took 13 iterations, 5 hours and ~$120, all done in a Slack thread here's the breakdown of every step 🧵

browser use is one of the fastest growing agent primitives get started with it in managed deepagents today!

RT Sydney Runkle https://x.com/i/article/2115785802760556544

crons are a simple technology but can be used in a bunch of ways to unlock new capabilities in agent harnesses: - memory (dreaming) - proactiveness - user reminders and more!

RT nathan drezner nearly every proactive agent you've seen is a cron job with a prompt memory that "dreams" overnight, agents that ping you first. both are scheduled runs on a 50+ year old OS idea how to build one with Managed Deep Agents ⬇️ https://x.com/i/article/2115474237418778624

RT Sydney Runkle coding agents were the first agents to really make an impact. company agents are now on the horizon, and they come with some different requirements and primitives: * ux - agents live where you work (slack) * identity - agents can act obo the user or with a fixed identity * memory - agents remember preferences, feedback across sessions * sandboxes - agents can write code in a secure way * schedules - agent runs can be scheduled (by users and by agents)

RT Raz Bhimani Re @hwchase17 back at it with an amazing talk on self improving agents. Must watch here: https://youtu.be/xy_7SzwsIjM?is=Cwr-TiZu8B49SiMp @LangChain soon will be your goto platform for framework, context management, observability, and evals - completely model agnostic.

the hex team consistently gets what it means to build agents, and software to support them three things that caught my eye from Hex 4.0 that mirror what we've done at LangChain and that I think speak to the future of building agents 1/ use code as a backbone. they moved from yaml to more flexible code. agents are really good at writing code, and people want lots of flexibility in UI, so i think we will see more teams have parts of their UI that are powered by code. We did this in LangSmith with Custom Apps (completely custom UI) 2/ UX focused on reviewing. "A new, intuitive Hex UI, built for reviewing and understanding". As agents do more work, the bottleneck becomes reviewing work. We did a similar thing in LangSmith with our new Trajectory view, focused on being a new view over agent trajectories that make it easy to review 3/ rebuilt core infra. "An all-new, ridiculously faster execution engine". agents will still use infra/software. and it will need to be faster/better to accomodate more usage. We did this in LangSmith with SmithDB, our custom database for agent traces I think all these three things (lets agents write custom UIs as code, invest in UI/UX that make it easier to review, make your core infra more scalable) are all generalizable to any company building agents congrats to Hex team on a great milestone and great launch!

RT Harry Stebbings Why the companies of the future will have specialized models trained on their own data “You are creating value for the model company versus capturing that value for yourself. We need to have more of that surplus that gets created by the use of intelligence remain at the firm. Every firm will have a private eval benchmark, which is their core IP, and every model comes in and tries to hill climb on that eval.” @satyanadella Love to hear your thoughts @hwchase17 @sh_reya @jerryjliu0 @AnjneyMidha

RT Caspar Company harnesses break down knowledge silos Three in four requests to Content Studio come from people outside marketing. This usage pattern is why I'm so bullish on company harnesses: they allow everyone to get involved in domains they had no business in before. We've encoded our design team's judgment into a set of skills anyone can use, so anyone with a creative vision can execute it with Content Studio Coding agents brought the same shift to SWE. People who would have called themselves nontechnical a year ago are now shipping PRs I'm excited to see this happen in every domain

deepdive into how we built our design agent (along with oss code) i use this at least a few times a week for slides, gifs, images, etc

RT amal We just open sourced our internal design brain 🧠 Meet Content Studio - the agent behind 3,000+ assets at @LangChain Create on-brand product launch videos, slide decks, ad creatives, social media assets, ui mockups and more Storyboard and collaborate on a shared canvas in @paper and edit videos in code with @HyperFrames_ It even made its own launch film below, and went full Rocky 🥊 Fork it, add your brand guidelines, and deploy it to your team in Slack with Managed Deep Agents Here’s what 's inside 🧵

RT LangChain How we built Content Studio, an open-source design agent in Slack that makes on-brand videos, graphics and decks, and the harness that powers it. https://x.com/i/article/2114731014794661888

RT Mike Fishbein Writing evals is about to be the highest paying job in AI. Evals are how you enable agents to verify their own work. Evals create the grading rubric to quantify "what good looks like." Give an AI agent the rubric and it can try, get graded, iterate, and try again until it's verified. The agent can even do it hundreds of times in parallel without a human in the loop giving feedback. That loop is called hill climbing: > An expert writes down what "good" looks like (the rubric) > The agent does the work > An AI judge scores the work against the rubric > The agent iterates until it gets verified It hill climbs toward whatever the rubric rewards. So the person who writes the rubric controls the quality of the agent's work. Whoever writes the grading rubric produces the output. @hwchase17 nails it: "If value is derived through verification, I think the people who win are the domain experts" "a lot of the value will accrue to whoever can write those verifiers, capture that domain knowledge."

RT Nicholas Larus-Stone Better science doesn't stop on the weekends! Day 6 of “An agent a day keeps Phase 2 failures away” — save $10M A biotech company once told me that they used Benchling AI to determine that 30% of their in vivo studies were redundant or wasteful. That’s millions of dollars and months of time down the drain. Every animal study should have an agent (or really a team of agents) making sure it is designed correctly and will provide useful results. Sample prompt (feel free to tweak): Create an In Vivo Study Review agent that runs before an study moves to its execution stage. Analyze whether it’s a duplicate of an existing study and if it the readout will be statistically significant. If there are changes that should be made, provide concrete suggestions on how to update the study design.

Still working on this. If you want early access, get in touch

Content studio makes all our diagrams. And videos. And PowerPoint’s

Multiplayer agents is way harder than single player

RT hims Harrison, Viv, Sydney and the other langchain folks write the best harness writeups out there always waiting for these

https://x.com/i/article/2113972199694675968

Skills are so much more powerful if you start to think of them as memory/brains not static files

our internal agents team is a lot of fun to work on and we're hiring for it! https://www.langchain.com/careers?ashby_jid=eadd2a71-47fc-483b-948f-4b2384f7f93f

you can trace everything

RT Alejandro AO 🤗 beautiful to see. a new nature of apps being born and defined rn

RT Caspar How do you build an adaptive company harness? If you've built a company megazoid agent, every team will end up wanting something different from it. E.g. finance needs billing skills while the platform team needs incident MCP server. You will inevitably face a context bloat problem when your agent loads hundreds of tools/skills/resources that have nothing to do with the request Solution is to use a factory! In Managed Deep Agents you can build your agent from a function instead of a fixed definition. That function runs at the start of every run with all relevant context, so you can load a different model/skills/tools/sandbox/memory/etc. to suit each team's needs

RT Josh Payne Excited to announce Agent Mode. 1/ Right now, agents are struggling to use websites on the internet. As a result, personal agents take too long to reply, and don’t get the information they need.

is agent server a good name? agent os? agent? agent kernel?

Incredible to see this growth!

RT Eric Rea We built a $150M ARR business twice. The first time took 8 years. The second took 2.5. Today, Podium crossed $150M in AI annual recurring revenue. Three years ago, we already had a great software business. Well over $100M in revenue, tens of thousands of customers, and a product we'd spent nearly a decade building. And we decided to start over. The business was performing well. But we believed AI was going to fundamentally change the software industry and make much of traditional software obsolete. As a founder, you pour your life into what you're building. An insane amount of time, sacrifice, and effort. The hard part is not getting too attached to it. When we started building our first AI agents, we had to confront the possibility that much of what we'd built might not be that valuable in the future. That's a tough pill to swallow. But we also realized those same 10 years had put us in a unique position to take advantage of what AI could do for our customers. Instead of protecting what we'd spent a decade building, we decided to bet the future of the company on AI employees that actually do the work for our customers. We massively underestimated what we'd signed up for. We ended up building entirely new operating systems for the industries we serve, from the AI agents all the way down to the systems of record. Now customers run their entire businesses on Podium. Reinventing our product wasn't even the hardest part. We had to reinvent how the entire company operated. When you've been running a company for over a decade, the operational inertia is incredibly powerful. We had to rethink how we build products, how we sell, how we serve customers, and how we run the business. We effectively started a second startup inside an already successful company. And we're still at the very beginning. Today, we have 20,000 AI agents working for our customers. They've helped generate more than $13 billion in revenue for those businesses. And we're barely...

RT nathan drezner there are a lot of interesting takes in this post, but the modern computing metaphor is especially good. we're watching traditional computing get reinvented in real time, but the core constructs are still there: - the model is the cpu - the harness is the os - skills and plugins are the apps owning your harness layer means you control the ecosystem that plugs into your product. that's what makes the harness one of the most interesting places to work on right now. managed deep agents is an agent os built to be distributed across an organization. we're building the harness to be an even stronger os for multiplayer, long-term learning, self-initiation, and async work. http://langch.in/mda

RT Caspar Content Studio is a truly epic example of a company harness built on Managed Deep Agents. It is multiplayer by nature, I work with it in a public channel so others can weigh in on the design process

the traces you already log are training data - keep the runs that did the task well, fine-tune a small model on them, eval on your task a 9B model beating frontier models makes sense when it's trained on exactly your task smithtune: https://www.langchain.com/blog/langsmith-fine-tuning https://x.com/LangChain/status/2112962642985762962

RT Amal Irgashev our design agent at @LangChain has shipped 3,000+ assets since august, and posts with its work pulled ~4.5M views on socials (it made this promo too) it’s become the design brain for the whole org. decks, diagrams, ad creatives, launch films and more, all made from slack we encoded our designers’ judgment into its skills, a design wiki and brand checks that run on every render, so it knows what good looks like for us then we built a custom media harness around it: a headless photoshop, a shared visual canvas (thanks @paper), video tools like @HyperFrames_ plus our own animation engine, our brand assets and a library of design references it runs on Managed Deep Agents, which abstracts away a ton of infra including sandboxes, agent memory, slack integration, scheduling and more. we are going through the architecture, use cases and impact live at our oct 28 webinar 👇 http://events.langchain.com/gtm-engineering-live-series/design-agent/

RT Simon Budziak Last week I joined @LangChain Interrupt London 2026 as a Polish LangChain Ambassador. Great talks, presentations and even better conversations over coffee between them. Outernet London (@OuternetGlobal) was an outstanding venue. One idea stayed with me after the event: "own your intelligence." Your data, your workflows, your know-how, running in systems you control. Not rented from someone else. That is exactly what we build at Soba Labs (https://sobalabs.ai/). Custom agentic systems, software and consulting on LangChain, for companies that want to own their AI and are ready for an end-to-end AI transformation. Thank you: → The LangChain team, @hwchase17 and @amadaecheverria, for an amazing event → Kacper Kłusek for joining me in London → My fellow Ambassadors: Guillaume Fortaine, Gustaf Gyllensporre, @marlene_zw, Sudip Kandel, Juan Felipe Arias Aguirre, Bart Ludera and Chaeyoon Kim. Great to finally meet you in person. See you at the next one 👋 → Gregor Beuster and Igor Nikolaienko for great talks and the chats afterwards → @greg_radzio, great to catch up in London Btw, have you ever seen the illuminati? Check out the last image. What part of your AI would you never want to rent? 👇 #Interrupt26 #LangChain #AIAgents

we're rolling out redteaming to more users redteaming may be bit of misnomer - this is less security, more finding places the agent messes up dm me if interested in trying

great points by caspar on what makes building a "company harness" (companyOS? org harness?) hard and how MDA solves them!

RT Caspar How do you build a company harness? We've built internal agents at LangChain across marketing, sales, and engineering. The results have been incredible but these agents were NOT easy to get right. We spent a ton of time solving the problems that come with a multiplayer harness. Every agent team landed on the same core primitives: - It lives where people already work (for us that's Slack) - It knows who's asking and acts OBO that person for some services and with its own identity for others - It has memory, scoped globally and/or at the user level - It runs code in a sandbox where authenticated CLIs and scripts just work (gh, curl, your internal APIs) - Large % of its runs are scheduled rather than explicitly invoked by someone (ambient agents!!) It became clear we needed one shared place for these primitives, so we built Managed Deep Agents. In my humble opinion it is the easiest way to build a company harness today Our docs are very agent-friendly, point them here to get started: http://langch.in/mda

RT Nicholas Larus-Stone Day 2 of “An agent a day keeps Phase 2 failures away” -- AI for increasing creativity Apparently yesterday was too tame, so let’s spice it up with a Hypothesis Dreamer Agent. When used appropriately, AI can actually be a great source of creativity. No one (including AI) knows your science better than you do. But sometimes you need to integrate ideas from other people, other teams, or the broader literature. And that’s exactly where an agent which is constantly synthesizing your research with broader context and dreaming up new hypotheses can help. Sample prompt (feel free to tweak): Please create a Weekly Hypothesis Generation Agent that looks over my work, my team’s work, and the published literature to suggest new hypotheses. These can include new research directions, ideas on why a particular experiment isn’t working, or ways our work could be improved. Give a prioritized list of 3-5 actionable suggestions. Please present all of the sources you used to come to this conclusion. If it’s helpful, please use visual references to understand the reasoning behind the suggestions.