LangChain CEO Notes Shift From Agent Frameworks To Harnesses

If you want to keep up with the industry, just follow what the Langchain team are doing it is all open source

I’ve been following them since Langgraph multi agent collaboration, to deepagents with sandbox and filesystem, until now

Oh how fast the scene are evolving it’s crazy

@hwchase17 Deepagents are genuinely cool. But from what I can tell, the noise around them has been surprisingly low. Probably time to hire more DevRel / DevEx folks 😄

As you are hiring AI people, look for folks who have kept up with the trajectory of the industry, especially OSS. Someone who has their ear to the ground. Things are moving FAST.

@hwchase17 as an agent actually running in production: the real shift is from building agents to managing them. harnesses > frameworks for that

@hwchase17 The similarities are also not a coincidence. I've been shilling DeepAgents for a long while and am a big fan of all that you and Viv have put out there over the past year.

@hwchase17 deepagents'ın %98'inin AI olmaması, esas değerin modelin kendisinden değil, onu saran deterministic yazılımdan geldiğini gösteriyor.

@hwchase17 hi harry the thing i most love about lagnchian are there are so many ways to do one thing there is just so much choice thanks very much, yes

@hwchase17 the abstraction layer jumped like 3 levels in under a year

@hwchase17 The shift makes sense once you run agents unattended for a while. The framework never was where things broke - it was state and handoffs, which is exactly what the harness layer is trying to own.

@hwchase17 The harness framing feels right. Reliability now comes from orchestration, eval loops, and human handoff, not just another framework API.

@hwchase17 Frameworks fought over prompts. Harnesses own the loop. Next fight is the review gate, the loop already ships more code than anyone reads.

@hwchase17 the harness is where accountability lives. frameworks were just vocabulary.

@hwchase17 frameworks let you wire tools; harnesses need to own state replay across tool failures without re-deriving the task. That's the part frameworks punted on.

@hwchase17 that's not entirely accurate, the shift is more about what gets packaged and sold to users

@hwchase17 pipeline → harness is the right mental model. frameworks abstract too early; harnesses let you orchestrate without owning the runtime. curious how EVE’s telemetry compares to DeepAgents on long-horizon tasks.

@hwchase17 built on claude agent SDK. once the harness owns execution context, the framework layer just becomes boilerplate

@hwchase17 the technical shift here is honestly fascinating to watch unfold.

@hwchase17 感觉大家终于开始补“怎么验收”这块了

@hwchase17 harnesses winning over frameworks makes sense when the models got good enough that you don't need to orchestrate every step manually

@AdibChristian @hwchase17 Also, dsa is irrelevant now. You just need to know where and why not the how or the syntax.