Mukil Loganathan

Head of Infrastructure @LangChainAI

[Harrison Chase @hwchase17]

Harness: ✅ (DeepAgents)
Sandboxes: ✅ (LangSmith Sandboxes)
Eval: ✅ (LangSmith Sandboxes
Model: integrate with all the popular models and providers

Plus we have the engine that helps you turn this flywheel: LangSmith Engine https://t.co/NLeUafWZtz

own your agents!

[Brace @BraceSproul]

Agent architecture is essentially a solved problem. What's not: memory. Great article by @jakebroekhuizen who's a true expert on this

[Mukil Loganathan @MukilLoganathan]

@ankush_gola11 @palashshah live mas.

[Mukil Loganathan @MukilLoganathan]

@brichall_ @LangChain We have something coming for this too!

[Mayor Zohran Kwame Mamdani @NYCMayor]

MSG requested a permit for a watch party for 500-999 fans. We approved that permit for 999 fans.

Mr. Dolan has now decided to cancel the watch party.

I know this is breaking hearts across our city.

But if there's one thing Knicks fans don't need permission for, it's showing up for our team wherever we may be — no matter the block or the borough.

Knicks in five.

[Viv @Vtrivedy10]

yo meek

I’ll be honest - if you want the top model for intelligence and are down spending any amount of bread on it, it’s the new Claude model called Fable 5 that dropped today (breaking all the benchmarks 🔥)

But there’s another piece of AI beyond a smart model —> it’s all the stuff you tell the model about yourself and your business + goals to help the model help you

That stuff is called the “context” and the stuff we put around the model to make it useful for our business (like tools to access your company info) is called the “harness”

These are all just fancy terms for:
1. we gotta tell the models what matters to us and how to do the job well

2. we gotta give models some tools to access the data and stuff we care about so it can go do our work for us

Another thing on AI for the masses

I know you’re all about access to the people - so tbh these really smart models are great but they’re mad expensive for the average person to use for their business

I think if you’re just getting into AI, then ChatGPT and Claude with the best models are great, but there’s this other family of Models called “Open Models” which means they’re fully available to any person to access and download for themselves and they run way cheaper

Some of their names are Arcee Trinity, Nvidia Nemotron, Kimi k2.6

Lots of people can unlock things in their business by using these way cheaper models to do the tasks they care about

I often say that most tasks don’t need some Einstein level intelligence to get shit done —> they need affordable, cheap, very smart intelligence that rocks as your business goals

I think the cheaper we can make AI for the world to access the better

No one should be priced out of access to smart ai because they don’t have the bread —> there’s a family of models called Open Models that’re way cheaper

Anyway you can prob just copy the above message to Claude and start having it explain stuff to you - the company i work at works on a ton of stuff to make models smarter for your use cases

The word you might come across is harness engineering

If you paste this message to Claude tell him I sent you and my DMs open if you need help cookin 🫡

[Mukil Loganathan @MukilLoganathan]

@MeekMill @Vtrivedy10 bring it home

[Leonard Tang @leonardtang_]

opportunity cost is insanely high for exceptional talent

[Samuel Spitz @samuel_spitz]

The best employees are the ones you don't have to manage

[tyler hogge @thogge]

Customers kingmake startups. Not VCs.

[Caspar Broekhuizen @caspar_br]

We're building this at LangChain

Fleet lets you create and manage a fleet of agents. Each agent specializes in a workflow, e.g. inbox management, blog writing, competitor research, candidate recruiting. These are Deep Agents with custom instructions, skills, tools, subagents, and memory. They continually improve with feedback. You can share them with your coworkers. You can configure them to run on a schedule. You can export their context files should you ever want to host them yourself

I think Fleet strikes a great balance: easy to use and still highly capable

We've put an inordinate amount of thought into the UX patterns that make that possible. For example, I love our 'channels' concept: you can configure your agent's communication channel (e.g. Slack, Teams, email, etc.) so it meets you where you work instead of forcing you into Fleet's UI

It's free to try out so give it a spin and share feedback: https://t.co/TRYcK32IBB

[Mukil Loganathan @MukilLoganathan]

wow havent played since season of ice on GBA but this is a movie

[Viv @Vtrivedy10]

the Spyro series rocks, beautiful worlds, fun story, one of my fave games growing up, so hype 💜

[Palash Shah @palashshah]

here's what it means to create a loop, and how we use them to automate our work

[Palash Shah @palashshah]

loops is just a fancy term for automated responses to signals. as an engineer, there's a set of signals in your workday that cause you to start streams of work.

someone messages you on slack about a bug, there's errors in your agent's tracing project, a customer leaves a bad review about your product.

[Mukil Loganathan @MukilLoganathan]

@sidpalas We support both bringing your own image and importing a dockerfile in Langsmith!
https://t.co/4XsDMdF8f6

[Harrison Chase @hwchase17]

Every platform will have a headless version

[LangChain @LangChain]

Sandboxes are already helping teams move from agents that answer questions to agents that can do work safely.

At @mondaydotcom, that means giving Sidekick a secure environment to write and run code for more advanced user workflows.

[LangChain @LangChain]

New in the LangSmith Sandboxes GA Release: Sandbox CLI

✅Build snapshots from Dockerfiles
✅Manage sandboxes
✅Open interactive consoles
✅Tunnel raw TCP
✅Use standard tools (ssh, scp, rsync, sftp) against a sandbox like any Linux box

https://t.co/QPoLcmtixW

[Mukil Loganathan @MukilLoganathan]

This was super fun! Check out LangSmith sandboxes, GA today!

[LangChain @LangChain]

.@MukilLoganathan’s Interrupt keynote on Sandboxes. https://t.co/oddQOs0Q6O

In 20 minutes, you’ll learn how to run agent code safely.

Isolated from your runtime, with network controls, persistent state, and snapshot/restore when things go wrong.

[LangChain @LangChain]

.@MukilLoganathan’s Interrupt keynote on Sandboxes. https://t.co/oddQOs0Q6O

In 20 minutes, you’ll learn how to run agent code safely.

Isolated from your runtime, with network controls, persistent state, and snapshot/restore when things go wrong.