Harrison Chase, co-founder and chief executive of LangChain, published an article on X on 24 October 2026 titled The five components of building a company harness, read on 24 October 2026. Chase writes that LangChain built internal agents “across marketing, sales, and engineering, and they have completely changed how those teams work.”
Chase says every team “landed on the same core primitives”, and lists five. A harness, in plain terms, is the software around an AI model: where it is reached, what it can touch, what it remembers and when it runs.
Business Pill · THE MODEL AND THE HARNESS
A short Business Engineer explainer on the difference between an AI model and the harness around it. It teaches the general idea only and says nothing about Chase’s article.
The key insight: As we read it, Chase’s list moves the hard part of company agents away from the model: four of Chase’s five components are about where an agent lives, whose identity it uses, what it remembers and when it runs.
The five components Harrison Chase, LangChain’s chief executive, lists for a ‘company harness’ in an X Article of 24 October 2026: it lives where people work (Slack), handles identity, keeps memory at two levels, runs code in a sandbox, and runs largely on schedules. Numbers are the article’s list order, not data.
Harrison Chase, LangChain’s co-founder and chief executive, shared an X Article on 24 October 2026. It sets out the five components Chase says LangChain’s internal agents share: they live in Slack, handle identity per tool, keep global and per-user memory, run code in a sandbox, and run largely on schedules. It is the primary source for this piece.
The Three Agents Behind It
Chase says the list comes from three internal agents. GTM Agent is “a proactive, always-on agent that assists with GTM activities”. OpenSWE, the engineering agent, “can work on and draft PRs, but also runs code review over all PRs”. Content Studio, the marketing agent, “can draft assets (blogs, images, PowerPoints, videos) in our style and with knowledge of our positioning”.
“All of these agents are built in the same way. We think this is a general recipe for building company harnesses,” Chase writes.
Identity Was the Hard Part
Chase calls identity “the thorniest and trickiest problem we tackled”: when an agent acts, on whose behalf does it act? Chase names two industry approaches: “On-behalf-of”, where the agent acts for the user who called it, and “Fixed credentials”, a set of credentials the agent uses no matter who is asking.
LangChain mixes both, Chase writes. OpenSWE opens pull requests with your GitHub handle when you message it, but in the background, as a code reviewer, it “acts with its own identity (GitHub app)”. GTM Agent uses a fixed identity for Salesforce and acts on the user’s behalf for email, and those email tools are only available in direct messages, not shared channels.
Memory, Sandboxes and Scheduled Runs
All three agents have two tiers of memory, Chase writes, global and per-user, and “Agents decide which facts they want to remember to which level of memory.” Chase adds: “This is probably the least explored area so far”.
On code, Chase says all three agents can write and execute it, connecting to a sandbox from outside, with a secret proxy that lets them reach services “in a safe and secure way from inside the sandbox itself.”
On scheduling, Chase writes that “A large percent of runs are scheduled or triggered in the background by events, rather than being directly kicked off by a user.” One example from the article: Chase asked GTM Agent for a daily brief, and it created its own cron job to send it.
Chase closes by saying all three agents are built on Managed Deep Agents, LangChain’s own product, which sits on its open-source Deep Agents harness, so the article also promotes what LangChain sells.
Connecting the Coverage
Chase’s list is about the parts of a harness. Our Anthropic subagents piece, published earlier today, reported Anthropic’s 22 October advice to fold specialist subagents back into one main agent loop, with one customer seeing a roughly 2x faster runtime.
Microsoft chief executive Satya Nadella wrote on 21 October that Microsoft runs a system where the harness, “including context, memory, and action space”, is separate from the model, as our Nadella harness piece reported. Today, our Rippling piece reported The Information’s account of Rippling moving most of its coding work from Anthropic to OpenAI models on price.
Anthropic donated its Agent Skills standard to the Agentic AI Foundation on 22 October, our Agent Skills piece reported. And on 18 October, our LangChain memory piece quoted Chase calling agent memory “super, super early”; in this article Chase again calls memory the least explored area.
What is new here: a named vendor laying out, from its own internal agents, the five parts it says every company harness needs, with identity singled out as the hardest.
The Structural Read
As we read it, identity is the component with the most detail in the article: Chase describes two approaches, on-behalf-of and fixed credentials, and says LangChain mixes them tool by tool and channel by channel.
As we read it, the scheduling point changes who starts the work: Chase writes that a large percent of runs are triggered by schedules or events rather than by a user, and that one agent created its own cron job when asked for a daily brief.
As we read it, the article is also a product description: Chase writes that all three agents are built on Managed Deep Agents, which LangChain offers, so the five components double as that product’s feature list.
Harrison Chase, ‘The five components of building a company harness’, 24 October 2026
“The thorniest and trickiest problem we tackled was identity, specifically: when these agents take actions, who do they take actions on behalf of?”
Three Implications
FOR TEAMS BUILDING INTERNAL AGENTS Chase’s article lists five parts to plan for: where the agent lives, identity, memory, sandboxed code and scheduled runs.
FOR SECURITY AND IT TEAMS Chase writes that some tools act with the user’s own login and others with a fixed account, and that LangChain limits some actions to direct messages or to whitelisted users.
FOR BUYERS COMPARING AGENT PLATFORMS Chase writes that LangChain’s agents run on its Managed Deep Agents product, which Chase says provides channels, crons, sandboxes, memory and identity.
The framework puts it as “The surface, not the model, governs autonomy. The agentic surface — the harness — is becoming where value pools in the entire AI stack.”
As we read it, Chase’s article is a field report from inside that surface: the five parts LangChain says its company harness needs, written by the vendor that sells one. The framework was published in July 2026 and does not discuss this article.
What Is Not Established
We read Chase’s article in full. It gives no figures on results, usage, costs or how many runs are scheduled; “Large %” is Chase’s phrase. We did not see LangChain data, test the agents or contact LangChain. The article describes LangChain’s own experience and product; whether the five components fit other companies is Chase’s view, not something we checked.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Harrison Chase says LangChain’s marketing, sales and engineering agents share five parts: they live in Slack, handle identity per tool, keep two tiers of memory, run code in sandboxes and run largely on schedules. Chase calls identity the hardest of the five.
A note on sourcing. This piece rests on Harrison Chase’s X Article of 24 October 2026, which we read in full. LangChain sells the product the article describes. We did not contact LangChain. Earlier figures come from our own reporting, linked in the text.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
