Harrison Chase(@hwchase17)

great insight into how companies are adopting and using open wiki in practice

Harrison Chase (@hwchase17) — Thu, 10 Sep 2026 20:01:27 GMT

great insight into how companies are adopting and using open wiki in practice

LangChain@LangChain

Credit Genie's coding agents used to guess at how a repo worked. Now they just check their OpenWiki.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬15🔄4❤️49👀7020📊18

⚡ Powered by xgo.ing

i expect computer use to get a lot better and more popular over the next few months! llms becoming ...

Harrison Chase (@hwchase17) — Wed, 09 Sep 2026 21:09:10 GMT

i expect computer use to get a lot better and more popular over the next few months!

llms becoming great at writing code is actually a big catalyst for this

Kyle Jeong@kylejeong

x.com/i/article/2097…

🔗 View Quoted Tweet

💬24🔄10❤️87👀12126📊32

⚡ Powered by xgo.ing

I like this framing a lot: agent improvement is harness improvement, not model improvement The inte...

Harrison Chase (@hwchase17) — Wed, 09 Sep 2026 16:15:11 GMT

I like this framing a lot: agent improvement is harness improvement, not model improvement

The interesting interventions are often at the tool boundary:
- what context you pass
- when tools are provided
- how you recover from failure
- what gets measured afterward

That’s the loop we’re building around deepagents (the orchestration logic) + LangSmith (how you measure)

Arky Yang@arkyyang

Beyond Prompts: Measuring and Optimizing LLM Tool-Agent Harnesses — new paper (arXiv, 4 Sep 2026; accepted EMNLP 2026). arxiv.org/abs/2609.05736

Treats harness optimization around a fixed model as budgeted selection: edits are guarded intercepts at the tool boundary, not rewrites of execution logic. PRISM clusters failures, routes each repair to a prompt, middleware, or joint surface, and picks candidates by gate pass-rate AND reliability (RelLift95), not score alone.

3 distinct takeaways for product builders:
- Report selected-harness reliability, not just mean lift. BH found the best individual harnesses but had negative reliable lift on both tau-bench domains — a strong explorer, an unreliable selector.
- Constrain middleware to tool-boundary edit patterns (silent correction, error blocking, prerequisite blocking). Unconstrained middleware (BH-MW) fell below its own prompt-only variant on Retail; the constraint, not the surface, is what makes middleware safely searchable.
- Route each failure to the right surface: prompts for global policy, middleware for locally-checkable execution invariants. Removing routing collapsed lift first-order on all three tasks; prompts alone miss runtime failures, unconstrained middleware adds selection risk.

Experiment setup (from the paper): fixed inner execution model gpt-5.4-mini, outer proposer claude-opus-4.7. Benchmarks: BFCL multi-round (600-case scorecard) and tau2-Retail / tau2-Telecom (74-case scorecard each of 114 tasks). Baselines BH, GEPA, MIPROv2 + PRISM; disjoint repair/gate/scorecard splits, 4 seeds, up to 16 runs. Baseline = native Deep Agents (v0.5.6). Metrics: mean held-out lift (pp), WorstLift, repeatability, RelLift95(B) by 5,000 bootstrap. PRISM-MW mean lifts 14.2 / 14.9 / 10.1 pp and positive RelLift95 on all three.

🔗 View Quoted Tweet

💬50🔄14❤️107👀11786📊55

⚡ Powered by xgo.ing

Open wiki is cool

Harrison Chase (@hwchase17) — Wed, 09 Sep 2026 01:15:44 GMT

Open wiki is cool

Zak Raicik@Zraic

I am OpenWiki's number 1 super fan @LangChain

🔗 View Quoted Tweet

💬4🔄2❤️19👀6570📊5

⚡ Powered by xgo.ing

memory comes built into deepagents!

Harrison Chase (@hwchase17) — Tue, 08 Sep 2026 22:08:33 GMT

memory comes built into deepagents!

Gipp 🦅@gippp69

holy sh*t. this is f**king insane.

langchain just documented the missing memory layer for agents that are supposed to improve after every run.

instead of stuffing everything into one giant prompt, deep agents can keep long-term memory, reusable skills and lessons from previous tasks.

[it takes a few minutes to understand the loop]

1/ run the task
2/ save what actually worked
3/ load that lesson into the next run

that’s how an agent stops starting from zero every single time.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬7🔄4❤️71👀16907📊19

⚡ Powered by xgo.ing

an increasingly hard part of agents is handling auth does the agent act as itself, or on behalf of ...

Harrison Chase (@hwchase17) — Tue, 08 Sep 2026 21:56:20 GMT

an increasingly hard part of agents is handling auth

does the agent act as itself, or on behalf of a user

we're trying to make this as easy as possible in managed deepagents!

Caspar@caspar_br

Agent auth is hard, but it shouldn't be!

Your agent needs to act as someone: sometimes that's one shared identity, sometimes it's the person asking. To get either right you'll need an OAuth app for each service, a token store, refresh logic, and a consent flow

Today we're fixing this with Managed Connections: define who your agent acts as in code, and the rest is handled for you:

𝚊𝚐𝚎𝚗𝚝 -> agent uses its own token
𝚞𝚜𝚎𝚛 -> agent uses the user's token, guiding them through the consent flow if they haven't connected yet

With a single argument you avoid the whole OAuth dance. Available now, in 𝚖𝚊𝚗𝚊𝚐𝚎𝚍-𝚍𝚎𝚎𝚙𝚊𝚐𝚎𝚗𝚝𝚜 0.7

🔗 View Quoted Tweet

💬28🔄17❤️120👀14111📊42

⚡ Powered by xgo.ing

harnesses should make context engineering easy "forking" subagents is a useful context engineering ...

Harrison Chase (@hwchase17) — Tue, 08 Sep 2026 19:43:33 GMT

harnesses should make context engineering easy

"forking" subagents is a useful context engineering trick that is now built into deepagents

LangChain OSS@LangChain_OSS

x.com/i/article/2097…

🔗 View Quoted Tweet

💬21🔄15❤️150👀27950📊44

⚡ Powered by xgo.ing

kai is very cool, and every company should have a kai this episode is a great resource on how strip...

Harrison Chase (@hwchase17) — Tue, 08 Sep 2026 16:06:14 GMT

kai is very cool, and every company should have a kai

this episode is a great resource on how stripe built it

here's another blog for some additional context: langchain.com/blog/how-strip…

claire vo 🖤@clairevo

The team at @stripe is setting the standard for internal AI platforms: minion coding agents, a custom prototyping rig, and now their company brain, Kai.

On today's episode of How I AI, Sharadh shows us how 1.5 engineers and 2 weeks got them a company brain, including: 
- projects as governance
- skill routing + telemetry
- a skills platform that works for 10k teammates

Plus, he and I debate the merits of gentle parenting your AI (esp when your company is running evals.)

Full episode on YT: youtube.com/watch?v=AbZODZ…

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬18🔄13❤️119👀27614📊47

⚡ Powered by xgo.ing

What are the tools you’d want to see in “open router for tools”

Harrison Chase (@hwchase17) — Sun, 06 Sep 2026 18:40:48 GMT

What are the tools you’d want to see in “open router for tools”

Fardeem@FardeemM

Openrouter for tools is a great idea. I don't wanna sign up for 1000 services

🔗 View Quoted Tweet

💬47🔄3❤️96👀30830📊48

⚡ Powered by xgo.ing

This is very similar to how we think about context management in Deep Agents: compaction should redu...

Harrison Chase (@hwchase17) — Sat, 05 Sep 2026 17:15:03 GMT

This is very similar to how we think about context management in Deep Agents: compaction should reduce what is sent to the model, not erase the underlying work history.

 In Deep Agents, the logic is:

 - every agent gets a filesystem to work with (can be real or virtual)
 - when a tool result is too large, we write the full result to the filesystem (/large_tool_results) and replace the model-visible message with a preview
 - when the conversation gets long, we offload older messages to the filesystem (/conversation_history) history first, then generate a summary
 - the agent can also decide to compact proactively via a compact_conversation tool, which uses the same offload + summarize logic

Summarization middleware with all this logic: github.com/langchain-ai/d…

OpenAI@OpenAI

This is GPT-6 Astra.

Anything you can do on a computer, Astra can do for you. Fast.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬62🔄27❤️239👀30553📊91

⚡ Powered by xgo.ing

smithdb is a database we built for agent trajectories from the ground up ankush has driven this fro...

Harrison Chase (@hwchase17) — Fri, 04 Sep 2026 18:09:37 GMT

smithdb is a database we built for agent trajectories from the ground up

ankush has driven this from 0->1, we're now hiring a lead for it

know anyone good?

Ankush Gola@ankush_gola11

We’re looking for someone to lead SmithDB at LangChain!

SmithDB is the database we’ve built from the ground up to power LangSmith. It’s optimized specifically for the access patterns that emerge when you’re storing and querying enormous volumes of agent traces.

It’s serving production traffic at significant scale with great performance, but it’s still early. The problems we are solving are genuinely fun and very hard: indexing, query execution, compaction, ingestion specifically for agent observability + making all of this fast and cost-efficient at massive scale.

We’re looking for an exceptional engineering leader to take ownership of this critical project. You’re likely a great fit if you’ve built database or distributed systems, led strong engineering teams through technically demanding projects, and have exceptional execution and project management skills.

Apply here or DM me directly: jobs.ashbyhq.com/langchain/d81c…

Learn more about SmithDB here: langchain.com/blog/introduci…

🔗 View Quoted Tweet

💬25🔄5❤️49👀12125📊23

⚡ Powered by xgo.ing

This is very true, and something we still try to do for all products (and even features) we launch. ...

Harrison Chase (@hwchase17) — Thu, 03 Sep 2026 16:47:28 GMT

This is very true, and something we still try to do for all products (and even features) we launch. Find a small group of customers who we think will push us the most on product and then build for them and make them happy

Jackson Dahl@jacksondahl

More on how early customers shape your business (@cursor_ai early work with Shopify, OpenAI, NVIDIA):

"Not all ARR early on is created equal. It's not about just that number. It's about the genetics.

I care far more that you've selected and are working with really thoughtful, avant-garde, demanding customers, than just your revenue... those are pressurizing functions for you to adapt your fitness."

@milesgrimshaw

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬8🔄6❤️34👀9247📊10

⚡ Powered by xgo.ing

Agent workspaces should be durable, inspectable, and swappable Deep Agents’ virtual file system is ...

Harrison Chase (@hwchase17) — Thu, 03 Sep 2026 16:17:41 GMT

Agent workspaces should be durable, inspectable, and swappable

Deep Agents’ virtual file system is intentionally behind a BackendProtocol, so the agent code can keep using read/write/glob/grep while teams choose the storage layer that fits production

Nice to see MongoDB support this!

MongoDB@MongoDB

MongoDB now powers the virtual file system behind @LangChain's Deep Agents. 🚀

This new integration lets you implement Deep Agents' BackendProtocol against MongoDB Atlas instead of building your own storage layer. And agent code calling read_file, write, glob, and grep don’t have to change.

Your retrieval and storage run through one backend instead of a maze of stitched-together tools. Vector, full-text, and hybrid search (via $rankFusion) route through Atlas; file reads and writes forward directly to S3.

For builders of long-running or multi-agent workflows: Plans, intermediate outputs, and knowledge artifacts persist across sessions, deployments, and sub-agent handoffs, so file-based context survives between runs instead of resetting each time.

🔗 View Quoted Tweet

💬28🔄7❤️58👀8047📊27

⚡ Powered by xgo.ing

we have an absolutely killer lineup for our Interrupt conference in NYC (sept 24th) in particular, ...

Harrison Chase (@hwchase17) — Wed, 02 Sep 2026 17:26:23 GMT

we have an absolutely killer lineup for our Interrupt conference in NYC (sept 24th)

in particular, im excited about the fireside chats we're doing with with founders at different parts of the agent stack:

- @alexatallah: openrouter is a key part of how companies maintain model optionality and try out the newest models. they've also been innovative on smart routing and model councils

- @bernhardsson: modal has one of the best DXs of any company i know, and plays a pivotal role at both the model inference layer as well as sandboxes (useful for coding agents, RL, and more)

- @GabeStengel: rogo is the leading ai partner for financial institutions, and is a great example of ai native app companies

in addition to these fireside chats, there are many more speakers (including langchain folks!) - so it will be informative and entertaining

come join us! interrupt.langchain.com/nyc

💬13🔄11❤️59👀8802📊20

⚡ Powered by xgo.ing

Agent traces shouldn’t only be useful to infra teams For most builders, the fastest way to debug an...

Harrison Chase (@hwchase17) — Wed, 02 Sep 2026 03:08:24 GMT

Agent traces shouldn’t only be useful to infra teams

For most builders, the fastest way to debug an agent is to replay the conversation + tool calls in the shape the agent experienced them

That’s the idea behind Messages View in LangSmith

Git Maxd@GitMaxd

The new LangSmith “Messages View”  dropped into my account (beta rollout)

Makes following @LangChain traces as easy as following the chat!

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬39🔄12❤️83👀12612📊36

⚡ Powered by xgo.ing

Deep research is a complex, multi-step search and synthesis workflow This is exactly the kind of wo...

Harrison Chase (@hwchase17) — Tue, 01 Sep 2026 16:13:22 GMT

Deep research is a complex, multi-step search and synthesis workflow

This is exactly the kind of workflow Deep Agents is meant to make easier to build

Sumanth@Sumanth_077

A research agent is only as good as the search layer behind it!

I built a Deep Research Agent using LangChain Deep Agents with a search-first architecture designed around grounded, source-backed research.

The agent plans the research in real time, breaks the work into smaller steps, searches across the web, reasons over the retrieved sources, and produces a fully cited answer.

For this project, I used Liner as the search and grounding layer.

Liner returns raw, structured search results with titles, URLs, descriptions, and dates. The agent then handles the reasoning and synthesis itself instead of relying on the search API to generate the final answer.

That separation matters.

Search is responsible for finding the right information. The agent is responsible for deciding what matters, comparing sources, connecting the evidence, and producing the final response.

The workflow looks like this:

Research → Plan → Search → Collect sources → Synthesize → Cite → Save report

Every claim in the final answer is backed by a source from the retrieved results, and the full report is saved as a downloadable Markdown artifact.

You can also inspect the entire run in the Deep Agents UI, including the research plan, search calls, retrieved sources, and final synthesis.

Liner handles retrieval, while the agent keeps control over reasoning, synthesis, and how the final answer is constructed.

Github Repo: github.com/Sumanth077/Han…

🔗 View Quoted Tweet

💬12🔄2❤️33👀8803📊14

⚡ Powered by xgo.ing

❤️ still early, lots to build, keep the feedback coming

Harrison Chase (@hwchase17) — Tue, 01 Sep 2026 02:57:09 GMT

❤️ still early, lots to build, keep the feedback coming

Shashwat Jain@JainAI29

I have been using langgraph for like a year and a half now and i can easily say the growth this company has had and the way they have listened to the community is quite remarkably unmatched. Awesome work @hwchase17

🔗 View Quoted Tweet

💬6🔄0❤️39👀8146📊7

⚡ Powered by xgo.ing

we're doing a meetup! "a summer evening on continual learning" partnering with our friends at prime...

Harrison Chase (@hwchase17) — Tue, 01 Sep 2026 00:30:02 GMT

we're doing a meetup! "a summer evening on continual learning"

partnering with our friends at prime intellect and baseten

come join!

luma.com/cwn8mze6

💬9🔄8❤️44👀15996📊16

⚡ Powered by xgo.ing

if only there was a company that helped you do both evals and harness engineering

Harrison Chase (@hwchase17) — Mon, 31 Aug 2026 19:38:36 GMT

if only there was a company that helped you do both evals and harness engineering

elvis@omarsar0

Next to evals, harness engineering is quickly becoming one of the most important skills for AI engineers to have today.

🔗 View Quoted Tweet

💬19🔄10❤️134👀19493📊28

⚡ Powered by xgo.ing

Call for agent infra

Harrison Chase (@hwchase17) — Mon, 31 Aug 2026 17:59:21 GMT

Call for agent infra

Alex@AlexatVester

Request for infra:

Who is going to build the open source version of the Codex-style in-app browser for agents.

Said differently - 
Who wants to be an overnight hero?!

@hwchase17@danshipper@davemorin

Call in the cavalry!

🔗 View Quoted Tweet

💬8🔄3❤️24👀8987📊10

⚡ Powered by xgo.ing

cost reconciliation at the trace level For AI apps, "how much did we spend?" is less useful than "w...

Harrison Chase (@hwchase17) — Mon, 31 Aug 2026 16:17:40 GMT

cost reconciliation at the trace level

For AI apps, "how much did we spend?" is less useful than "which workflow, tool call, prompt path, or retry pattern drove that spend?"

LangChain@LangChain

Clay uses LangSmith to reconcile 99.5% of its AI spend. Head of AI @jeffbarg on why cost observability finally clicked once LangSmith was in place.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬17🔄4❤️47👀11856📊18

⚡ Powered by xgo.ing

Separation of model and harness One of the first commandments of agents building; don’t get locked ...

Harrison Chase (@hwchase17) — Sat, 29 Aug 2026 22:09:36 GMT

Separation of model and harness

One of the first commandments of agents building; don’t get locked in

💬53🔄13❤️133👀18514📊65

⚡ Powered by xgo.ing

the original and internal name for LangSmith Insights was CLIO, which was Anthropics name for the re...

Harrison Chase (@hwchase17) — Sat, 29 Aug 2026 20:18:17 GMT

the original and internal name for LangSmith Insights was CLIO, which was Anthropics name for the research they did that inspired us to build Insights

now Anthropic has rebranded CLIO as Anthropic Insights

🤝
docs.langchain.com/langsmith/insi…J

Saffron Huang@saffronhuang

Hi twitter, here is a lil explainer of Anthropic Insights, which is how we analyze usage data in a privacy-protective way.

We just launched an exciting pilot enabling independent researchers to run studies with it (t.co/suLJcARAu1) and have understandably gotten questions about how it works without exposing individual chats. (also generally a good time to think about how third party assessments might work!)

Anthropic Insights lets us distill population-level patterns without inspecting individual conversations, so that we can equip the public/policymakers with info on how AI is changing life and work. It leans on Claude to provide multiple layers of privacy safeguards -- 

1) First we take a random sample of conversations, and Claude removes names and personal details.

2) Claude reduces each convo to a handful of specified attributes, like topic, language, and length.

3) Claude creates groupings of similar attributes and generates descriptions of those groups to create broad, population-level stats. E.g. a topic group might read: "Professional email drafting" or "Create, automate, and publish social media content"

4) Narrow themes get filtered out -- clusters with too few conversations / from too few accounts get removed.

Basically it's like Google Trends but for Claude usage, showing aggregated usage patterns that we can see change over time.

This is the system we use to study societal & economic impacts, affective use, and more at Anthropic. It’s also now how we can enable independent research on AI’s impacts outside the lab. 

With the independent research pilot we also did a privacy audit with a group of AI security researchers from Imperial College, who were unable to re-identify any users.

More technical details about Anthropic Insights here: t.co/X82ttb7f4b + happy to answer additional q's.

🔗 View Quoted Tweet

💬4🔄5❤️35👀9791📊8

⚡ Powered by xgo.ing

“Pair this with the right observability” Langsmith 🔥

Harrison Chase (@hwchase17) — Sat, 29 Aug 2026 04:50:31 GMT

“Pair this with the right observability”

Langsmith 🔥

Prakalp Choubey@ChoubeyPrakalp

@hwchase17 Pair this with the right observability and the system is unstoppable

🔗 View Quoted Tweet

💬5🔄2❤️18👀9106📊6

⚡ Powered by xgo.ing

Models labs will create great harnesses and ecosystems for their models but will block model access ...

Harrison Chase (@hwchase17) — Sat, 29 Aug 2026 02:16:14 GMT

Models labs will create great harnesses and ecosystems for their models but will block model access to harnesses of other labs

Only choice for a harness that works across models is one not associated with a lab

Long live LangChain

OpenAI@OpenAI

We’re ending our partnership with Cursor following its acquisition by SpaceX. Under our proposal, Cursor’s direct access to our models would end on November 12.

We know that the people most affected by this decision are the developers who rely on OpenAI models in Cursor. We care about their experience in this transition and we’re ready to go above and beyond to support them.

openai.com/index/our-deci…

🔗 View Quoted Tweet

💬56🔄21❤️340👀48509📊70

⚡ Powered by xgo.ing

Follow @colifran_ for all things wiki! Wikis are a great way to represent knowledge in a simple and...

Harrison Chase (@hwchase17) — Fri, 28 Aug 2026 23:20:53 GMT

Follow @colifran_ for all things wiki!

Wikis are a great way to represent knowledge in a simple and easy to understand way

george salapa 🜁@jurajsalapa

seems remarkably similar to langchain wiki?
also interesting how everything is collapsing to some shape of a loop and instruction file

🔗 View Quoted Tweet

💬8🔄12❤️104👀18421📊34

⚡ Powered by xgo.ing

Would love feedback on our support for the new mcp spec!

Harrison Chase (@hwchase17) — Fri, 28 Aug 2026 19:52:45 GMT

Would love feedback on our support for the new mcp spec!

Sydney Runkle@sydneyrunkle

we're adding support for the new MCP spec in @LangChain open source!

the new API is built on top of FastMCP, so you can take advantage of their excellent devx for building MCP servers and clients.

an early version is available in langchain==1.4.0a2; let us know what you think!

github.com/langchain-ai/l…

🔗 View Quoted Tweet

💬11🔄2❤️38👀11673📊11

⚡ Powered by xgo.ing

this is why benchling will create the best ai agent for science, not ant/oai strongly believe in a ...

Harrison Chase (@hwchase17) — Thu, 27 Aug 2026 21:09:33 GMT

this is why benchling will create the best ai agent for science, not ant/oai

strongly believe in a many-agent world, each one domain specific

Nicholas Larus-Stone@nlarusstone

x.com/i/article/2092…

🔗 View Quoted Tweet

💬16🔄7❤️61👀12982📊23

⚡ Powered by xgo.ing

openwiki now supports okf 2.0

Harrison Chase (@hwchase17) — Thu, 27 Aug 2026 17:34:14 GMT

openwiki now supports okf 2.0

Colin Francis@colifran_

the claims runtime in openwiki v0.4.0 gives your wiki a persistent source of truth to help it forget and self-correct.

but how would a reader know:
1⃣ what sources a page was built from?
2⃣ if a page is fully verified?
3⃣ who generated the page in the first place?

to answer that we migrated openwiki to support OKF v0.2

the claims runtime powers page-level sources and verification and openwiki separately stamps who generated the page and when.

claims=content-level truth
okf=page-level trust

special thank you to @seo_dev2142 for kicking off okf v0.2 support in openwiki!

🔗 View Quoted Tweet

💬8🔄7❤️63👀10346📊18

⚡ Powered by xgo.ing

How we think about evaluating wikis Including whether wikis are actually even helpful!

Harrison Chase (@hwchase17) — Wed, 26 Aug 2026 15:44:24 GMT

How we think about evaluating wikis

Including whether wikis are actually even helpful!

LangChain@LangChain

x.com/i/article/2092…

🔗 View Quoted Tweet

💬16🔄7❤️94👀17486📊31

⚡ Powered by xgo.ing

Deepagents is becoming a multiplayer harness Auth, memory, etc If you’re building an agent that y...

Harrison Chase (@hwchase17) — Wed, 26 Aug 2026 02:47:14 GMT

Deepagents is becoming a multiplayer harness 

Auth, memory, etc

If you’re building an agent that you want to expose to multiple users in same thread and are thinking about these issues - reach out! Would love to chat

💬39🔄5❤️101👀14345📊50

⚡ Powered by xgo.ing

its relatively easy to generate a wiki the first time, but how do you update them reliably? openwik...

Harrison Chase (@hwchase17) — Tue, 25 Aug 2026 16:20:03 GMT

its relatively easy to generate a wiki the first time, but how do you update them reliably?

openwiki v0.4.0 improves in this dimension - it "forgets" better

Colin Francis@colifran_

openwiki v0.4.0 just went live 🎉

the focus? forgetting, accessibility, trust

1⃣ stale docs detection with forgetting and self-correction
2⃣ drive init and update through claude, codex, and opencode
3⃣ okf v0.2 support for provenance and verification

🧵1/4

🔗 View Quoted Tweet

💬16🔄7❤️29👀7302📊15

⚡ Powered by xgo.ing

trying to think about how to most easily create evals we launched a skill to help with this iterati...

Harrison Chase (@hwchase17) — Tue, 25 Aug 2026 15:09:43 GMT

trying to think about how to most easily create evals

we launched a skill to help with this iterative process

Viv@Vtrivedy10

x.com/i/article/2092…

🔗 View Quoted Tweet

💬18🔄20❤️233👀35661📊64

⚡ Powered by xgo.ing

Evals matter

Harrison Chase (@hwchase17) — Mon, 24 Aug 2026 17:30:09 GMT

Evals matter

Brendan (can/do)@BrendanFoody

OpenRouter is an LLM gateway, not a router.

Some of the investment in routers may be missing the plot.

80% of the work in building a good router is building a good eval. The routing logic is the easy part.

🔗 View Quoted Tweet

💬17🔄4❤️106👀33779📊26

⚡ Powered by xgo.ing

A hard part of agents is connections - seems easy but hard to get right/seamless Managed deepagents...

Harrison Chase (@hwchase17) — Mon, 24 Aug 2026 17:29:21 GMT

A hard part of agents is connections - seems easy but hard to get right/seamless

Managed deepagents tackles it for you 

More to come!

Christian Bromann@bromann

Starting today, a @LangChain's Managed Deep Agent deploy provisions your Slack app for you 🤯

No manifest. No OAuth redirects. No bot tokens to copy around.

One command, and your agent says 👋 in Slack.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬11🔄2❤️35👀10776📊13

⚡ Powered by xgo.ing

Open bot!

Harrison Chase (@hwchase17) — Thu, 20 Aug 2026 18:37:36 GMT

Open bot!

Atai Barkai@ataiiam

🎉 Introducing 𝙾𝚙𝚎𝚗 𝙱𝚘𝚝

An open source Grok Bot that works with ANY agent harness, designed for real companies.

It includes:

- AI Coworkers
- Generative UI
- Computer use (remote/local)
- Agent-human handoffs
- Full data recording, owned by you
github.com/CopilotKit/ope…je0KRts5

We're using this internally at @CopilotKit and it's changing the way we work forever.

Powered by CopilotKit and AG-UI.

More info below 👇

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬13🔄25❤️367👀66998📊79

⚡ Powered by xgo.ing

very cool launch we did a webinar with jeff on "wiki" style memory and it's clear he'd thought abou...

Harrison Chase (@hwchase17) — Thu, 20 Aug 2026 17:47:22 GMT

very cool launch

we did a webinar with jeff on "wiki" style memory and it's clear he'd thought about this problem a lot

(webinar here: youtube.com/watch?v=Lsut4T…)

Jeff Huber@jeffreyhuber

I’ve been looking forward to today for 3 years 

Today we’re announcing Foundation - Chroma’s solution to memory 

Our research preview of this technology builds self-improving memory from your agent sessions. Try it out at trychroma.com/foundation

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬11🔄5❤️63👀24618📊22

⚡ Powered by xgo.ing

lots of observability & evals platforms not a lot of platforms that help close the loop and hav...

Harrison Chase (@hwchase17) — Thu, 20 Aug 2026 00:26:00 GMT

lots of observability & evals platforms

not a lot of platforms that help close the loop and have an agent that suggests fixes, adds evals itself, etc

YOЯNOC@conroywhitney

@hwchase17 BTW your NYC ads worked 😂 just learned that's why we chose LangSmith

🔗 View Quoted Tweet

💬18🔄7❤️38👀11033📊16

⚡ Powered by xgo.ing

top tier webinar tmrw! @Vtrivedy10 (@LangChain) @willcb (@PrimeIntellect) @AEllisBloor (@baseten) a...

Harrison Chase (@hwchase17) — Wed, 19 Aug 2026 17:50:54 GMT

top tier webinar tmrw!

@Vtrivedy10 (@LangChain) @willcb (@PrimeIntellect) @AEllisBloor (@baseten) and I will will be jamming on how to automate more parts of the agent improvement loop (eval and environment engineering in particular)

Come join us: events.langchain.com/webinar/Toward…

💬10🔄13❤️70👀25462📊22

⚡ Powered by xgo.ing

+1 to this part of this is due to coding agent standards - agents.md and skills are just markdown f...

Harrison Chase (@hwchase17) — Wed, 19 Aug 2026 17:22:13 GMT

+1 to this

part of this is due to coding agent standards - agents.md and skills are just markdown files/directories!

Sergio R.@TheRealSergio91

agents are starting to look less like apps and more like directories

instructions. skills. tools. memory. identity. channels. schedules. evals.

the harness + runtime become infra underneath it all. really like where this abstraction is going

🔗 View Quoted Tweet

💬13🔄3❤️40👀8465📊15

⚡ Powered by xgo.ing

🎓New YouTube playlist: Managed Deep Agents Gives an overview of Managed Deep Agents, and then each...

Harrison Chase (@hwchase17) — Wed, 19 Aug 2026 17:14:27 GMT

🎓New YouTube playlist: Managed Deep Agents

Gives an overview of Managed Deep Agents, and then each video dives deep into core concepts.

Launching with six videos!

1⃣ Intro:youtu.be/xdrB53bgpp0D
2⃣ Conceptual Overview:youtu.be/TUJmfeGTr1Qx
3⃣ Quickstart:youtu.be/L54dR9qMKzch
4⃣ Instructions and Context Hub:youtu.be/8HIoZV7qlwAr
5⃣ Skills:youtu.be/Lhru0yMI2as8
6⃣ Tools:youtu.be/piBwvWgHqkY1

Playlist link:youtube.com/playlist?list=…B

💬4🔄37❤️185👀17817📊66

⚡ Powered by xgo.ing

new onboarding for managed deep agents

Harrison Chase (@hwchase17) — Tue, 18 Aug 2026 22:54:52 GMT

new onboarding for managed deep agents

Caspar@caspar_br

mda 0.5.3 is a thing of beauty

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬6🔄8❤️60👀12984📊15

⚡ Powered by xgo.ing

I love working with Sydney and team and you will too! Come join us

Harrison Chase (@hwchase17) — Tue, 18 Aug 2026 22:04:26 GMT

I love working with Sydney and team and you will too!

Come join us

Sydney Runkle@sydneyrunkle

we're hiring open source devs @LangChain

looking for people who are building at the frontier of agents and are excited to quickly develop ownership

apply: langchain.com/careers?ashby_…

🔗 View Quoted Tweet

💬9🔄4❤️37👀13476📊10

⚡ Powered by xgo.ing

Slack is the best ux

Harrison Chase (@hwchase17) — Tue, 18 Aug 2026 19:52:39 GMT

Slack is the best ux

verrsane@verrsane

@hwchase17 if you dont have ur agents working overtime in slack/teams , ngmi

🔗 View Quoted Tweet

💬6🔄0❤️15👀6150📊6

⚡ Powered by xgo.ing

channels are how you interact with your managed deepagents eg slack pretty diagram 👇

Harrison Chase (@hwchase17) — Tue, 18 Aug 2026 17:37:53 GMT

channels are how you interact with your managed deepagents

eg slack

pretty diagram 👇

💬10🔄5❤️61👀11487📊22

⚡ Powered by xgo.ing

🚀Today we launched LangSmith Tuned Evaluators, starting with Perceived Error. Tuned Evaluators run...

Harrison Chase (@hwchase17) — Tue, 18 Aug 2026 16:45:21 GMT

🚀Today we launched LangSmith Tuned Evaluators, starting with Perceived Error.

Tuned Evaluators run on production traces to catch undesirable agent behavior and attach feedback that you can use in your agent improvement processes.

In our benchmark, our tuned model beat frontier models at 82% lower cost.
langchain.com/blog/introduci…Y

💬19🔄18❤️121👀20597📊41

⚡ Powered by xgo.ing

totally agree! here's how we architected deepagents to enable this deepagents runs connected to a "...

Harrison Chase (@hwchase17) — Sun, 16 Aug 2026 16:38:33 GMT

totally agree! here's how we architected deepagents to enable this

deepagents runs connected to a "backend". this backend needs to expose filesystem like operations, but it does not have to be a filesystem. it could be a database, object storage, or a real filesystem - it just has to expose read/write/edit etc operations

this backend could also be what we call a "sandbox". if a sandbox, it needs to expose an "execute" command which lets it execute code

this backend is SEPARATE from where the agent loop runs. this allows us to "separate the brains from the hands" (anthropic.com/engineering/ma…)

deepagents is built on top of langgraph, which means we can easily deploy it with MCP, a2a, and other standard endpoints

we use this architecture to power many different types of experiences

first, we can create a classic TUI like coding experience. we do this by giving deepagents a "sandbox" that is running locally in the same directory; deloying deepagents locally behind a light weight server; and then connecting to it with the TUI acting like a frontend. see dcode for an example of this docs.langchain.com/oss/deepagents…

second, we can create a cloud coding experience. we can do this by running deepagents on LangSmith deployments for a production scale deployment, and connecting to a sandbox running on modal, daytona, e2b that is running elsewhere. we can then build a frontend to connect to langsmith deployments and let users interract with it there, and also expose it in slack to let users interract with it there. note: both slack and web ui connect to the same backend, so you can switch between them seamlessly. code: github.com/langchain-ai/o…

of course - deepagents can be used to create agents that are NOT coding agents. a lot of agents still need to write and execute code, so this architecture is still very useful. but for some the code execution is overkill, and thats where you can swap to a "fake" backend, and still let it have the ability to interract with files (good for context engineering!) without having to spin up a full sandbox. 

for a really easy way to create these types of agents - see managed deepagents: langch.in/u8s5cgu

Patrick Collison@patrickc

I love agentic coding harnesses, but they shouldn't be primarily terminal-based. The terminal is great for quick and precise commands, but information density is extremely low and UI affordances are minimal. Maybe provision of TUIs is worthwhile for occasional use (when establishing a tunnel is too annoying, or something), but it feels very strange for this to be the default modality. It took a long time for dynamic language REPLs to break out of the terminal (Jupyter notebooks and similar); I hope we don't have to wait as long for the harnesses.

🔗 View Quoted Tweet

💬39🔄15❤️216👀31839📊78

⚡ Powered by xgo.ing

gave a talk "owning your intelligence" - ty @sequoia @sonyatweetybird for having me talked about ha...

Harrison Chase (@hwchase17) — Sat, 15 Aug 2026 15:45:42 GMT

gave a talk "owning your intelligence" - ty @sequoia@sonyatweetybird for having me

talked about harnesses and evals and the role they play in owning your intelligence

TLDR:
> agents = model + harness + context
> model - own the weights using something like @FireworksAI_HQ 
> context - memory needs to be portable
> harness - needs to be model agnostic. also needs to be good at bringing right context to llm. "right" context may depend on your use case, which is why an open/configurable harness helps
> how to use middleware in langchain/deepagents to configure your harness
> how to use langgraph to fully own your cognitive architecture
> why evals/obs matters - some quotes from @satyanadella 
- “Create your private evals, because evals define what “good” looks like inside the organization”
- “retain ownership of your organization’s memory, traces, feedbacks, decisions, and institutional context”
- “you create your own continuous learning loop (i.e. hill climbing machine) that will allow your AI investments to compound the value of your firm”
> how to use harbor for evals
> tracing is important
> evals + observability only matter so you can set up a data flywheel
> data flywheel = run agent -> collect traces -> find interesting traces -> use those to improve
> demo of langsmith engine which does exactly this!

full video: youtube.com/watch?v=HI2q3c…

💬34🔄26❤️250👀33559📊97

⚡ Powered by xgo.ing

It’s beautiful

Harrison Chase (@hwchase17) — Fri, 14 Aug 2026 20:14:24 GMT

It’s beautiful

Caspar Broekhuizen@caspar_br

anatomy of a managed deep agent. langch.in/mda

🔗 View Quoted Tweet

💬11🔄119❤️1615👀246619📊450

⚡ Powered by xgo.ing

some great new LangSmith docs on traces vs threads vs trajectories (new concept) observability data...

Harrison Chase (@hwchase17) — Fri, 14 Aug 2026 19:11:10 GMT

some great new LangSmith docs on traces vs threads vs trajectories (new concept)

observability data is no longer just for observability - its also for memory & learning

having a really clear mental model of this data is incredibly helpful!

https://t.co/UgPfRcVwZg

💬10🔄17❤️119👀22734📊41