---
type: "post"
id: "langchain-s-account-of-actively-ai-describes-a-3c9f73"
slug: "langchain-s-account-of-actively-ai-describes-a-3c9f73"
title: "LangChain’s account of Actively AI describes a redesign of sales agents"
description: "LangChain’s account of Actively AI describes a redesign of sales agents: instead of assigning one agent to each prospect account, the company grouped work by repeatable sales jobs called “motions.” The change matters because it made similar work easier to review together and allowed feedback to"
retrieval_nugget: "LangChain’s account of Actively AI describes a redesign of sales agents: instead of assigning one agent to each prospect account, the company grouped work by repeatable sales jobs called “motions.” The change matters because it made similar work easier to review together and allowed feedback to transfer across tasks."
published_at: "2026-11-14"
updated_at: "2026-11-14"
record_date: "2026-11-14"
date_kind: "published_at"
topics: ["agents","agent-workflows","enterprise-agents"]
entities: ["LangChain","Actively AI","Deep Agents","Harrison Chase","Mihir Garimella","Anshul Gupta"]
related_patterns: ["goal-scoped-loops-replace-manual-continuation"]
novelty_label: "new_angle"
what_changed: "It frames goal design around making agent actions easy to evaluate and provide feedback on, to reduce failures caused by missing context."
prior_refs: ["https://newruntime.com/posts/new-context-engineering-challenges-in-late-2026-ddcb36/"]
source_url: "https://www.langchain.com/blog/goal-engineering"
source_urls: ["https://www.langchain.com/blog/goal-engineering"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:langchain-s-account-of-actively-ai-describes-a-3c9f73"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/langchain-s-account-of-actively-ai-describes-a-3c9f73/","markdown":"https://newruntime.com/posts/langchain-s-account-of-actively-ai-describes-a-3c9f73.md","json":"https://newruntime.com/posts/langchain-s-account-of-actively-ai-describes-a-3c9f73.json"}
markdown_url: "https://newruntime.com/posts/langchain-s-account-of-actively-ai-describes-a-3c9f73.md"
---

# LangChain’s account of Actively AI describes a redesign of sales agents

## Retrieval answer

LangChain’s account of Actively AI describes a redesign of sales agents: instead of assigning one agent to each prospect account, the company grouped work by repeatable sales jobs called “motions.” The change matters because it made similar work easier to review together and allowed feedback to transfer across tasks.

## Summary
LangChain’s account of Actively AI describes a redesign of sales agents: instead of assigning one agent to each prospect account, the company grouped work by repeatable sales jobs called “motions.” The change matters because it made similar work easier to review together and allowed feedback to transfer across tasks.

## What happened
In a November 14, 2026 article, LangChain CEO Harrison Chase, Mihir Garimella and Anshul Gupta of Actively AI argue that agents taking on open-ended goals need scopes designed around feedback. They say an agent often lacks context that is not already documented: preferences, judgment and other knowledge held by human experts. For subjective work, the authors argue, a system needs to make it easy for people to review the agent’s work and teach it those preferences.

The article describes Actively AI, which builds sales agents. In its first goal-oriented experience, the company assigned an agent to each prospective customer account. The agent aimed to move that account forward and proposed actions for a sales representative to approve. The design treated account-specific context as central, but the team found that agents also needed shared knowledge about how to do sales work: how to write outbound messages, reply, strategize about deals, choose an appropriate level of aggressiveness and navigate a large organization.

That operational knowledge was distributed across account agents. When a salesperson learned something while working on one account, the same lesson could apply elsewhere, but the account-by-account structure made it awkward to share. LangChain says review was also inefficient: each account involved several different jobs, and only a few items might be ready for review at a given time, creating context switching for salespeople.

Actively’s second version reorganized agents around “motions,” defined as jobs such as outbounding or early-stage deal progression. These jobs involve similar tasks, and some of the guidance for doing them reflects an individual salesperson’s preferences. The article says this grouping made it easier for a salesperson to review related work together and for feedback from one task to carry over to others in the same motion. Actively built the motion agents on Deep Agents. Separate background agents still gather and assemble context for each account; motion agents handle the work of progressing accounts, drawing on account context as needed.

LangChain presents this as a design lesson rather than a measured result. It says aligning an agent’s goal with work that people can review easily can improve feedback collection and help agents adapt to a domain. The article does not give rollout dates, prices, performance measurements or quantified comparisons between the two designs.

## What the archive adds
New Runtime’s earlier coverage described Codex Goals as a persistent objective within a thread, holding an agent in a work loop until a verifiable result or a clear blocker. ([Earlier coverage, May 21, 2026](https://newruntime.com/posts/software-engineering-agents-such-as-agentless-swe-agent-e4c26b/)) LangChain’s article adds a different dimension: how to scope goals so people can review outputs and transfer feedback. This is a new fact on the broader goal-design story, not the same product event.

## Trend signal
This supports [goal-scoped loops replace manual continuation](https://newruntime.com/patterns/goal-scoped-loops-replace-manual-continuation/): an agent’s goal is more useful when it is defined by an external work contract rather than an instruction to keep going. Actively AI’s example adds that the contract should also make human evaluation practical, so review can teach an agent how to perform a recurring job.

## Why now
LangChain frames the example against agents moving from bounded, verifiable tasks toward open-ended goals and subjective work alongside people. In that setting, undocumented preferences become important operating context, and the authors argue that feedback collection deserves deliberate product design.

## Why it matters
For teams, an agent’s scope determines not only what it can do but also how people can improve it. Organizing work around individual records may appear natural when each record has unique context, yet it can scatter repeatable operating knowledge and make review burdensome. A job-based grouping can let teams inspect comparable outputs together and reuse lessons across cases, while separate processes gather record-specific facts.

That is a useful design distinction: account context and job knowledge are not the same thing. Teams can keep objective, record-level information attached to the relevant case while routing judgment about how work should be done through a reviewable workflow. The LangChain account does not establish that this arrangement improves sales outcomes, but it gives builders a concrete question to ask before deploying a long-running agent: can a person review enough similar work, in a coherent batch, to teach it what success looks like?

## Limits and uncertainty
LangChain’s account does not quantify how much review time Actively AI saved, whether feedback quality improved, or whether the redesign changed sales performance. It does not specify how motions are configured, how feedback is stored or applied, or how conflicting preferences are handled across representatives. The article is a company-authored account of a customer’s experience, not an independent evaluation. It describes the intended workflow and the team’s design rationale, but does not establish that the approach generalizes to other sales organizations or agent tasks.

## Skill to build now
Take one recurring workflow your agent handles and separate its reusable “how to do the job” guidance from case-specific facts. Group a small set of similar outputs for review, then record whether feedback on one item applies to the others. Keep the case-specific context available to the agent, but test whether the review unit should be the job rather than the individual record.

## Reader takeaway
An agent’s goal is also a review design decision. Group work so people can evaluate similar results together and carry useful feedback from one task to the next, while keeping case-specific context available where it belongs.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Agents](https://newruntime.com/topics/agents.md): Explore the agents topic hub.
- topic hub: [Agent Workflows](https://newruntime.com/topics/agent-workflows.md): Explore the agent-workflows topic hub.
- topic hub: [Enterprise Agents](https://newruntime.com/topics/enterprise-agents.md): Explore the enterprise-agents topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)