LangChain published a strategy essay by Harrison Chase on July 25, 2026, arguing that companies will not build lasting advantage on generic AI alone. The post, titled "What does it mean to 'own your intelligence'?," runs an estimated 9 min and appears in Chase's "Harrison's In the Loop" column on the LangChain Blog.
Chase opens with a prediction: over the next five years, every company will use AI in one of two ways, either to run critical parts of their business or as part of the product they sell to customers. In both cases, he writes, generic intelligence will not be enough.
The essay defines ownership as control over the parts that determine how intelligence behaves, how it is managed, and whether it compounds over time. Chase is explicit that this does not mean building every layer from scratch. Off-the-shelf AI is general and useful for common questions and broad reasoning, he writes, but to assume that generic intelligence can run operations inside a company would be a gross simplification.
Why Generic Models Fall Short
The argument rests on operational specifics. Companies have many details that matter more as AI is embedded deeper into their work, Chase writes, and those specifics will not be in a generic model's weights.
He uses an insurer's claims-processing operation as the example. A generic model can read policy documents, summarize claims, and explain common coverage concepts. But actual claim handling depends on the insurer's own policy language, state-by-state regulatory requirements, fraud signals, historical claim patterns, escalation rules, customer tiering, and risk tolerance. A generic model knows the meaning of "deductible," Chase writes, but not how to handle a specific claim type for a specific customer, jurisdiction, policy, and evidence.
The same logic applies to vertical AI startups building support agents, legal assistants, or coding agents. For those companies, the base model is not the product. The product is the system around the model: workflows, retrieved context, tools, evals that define quality, memory, and a feedback loop that improves with each customer interaction. The advantage, Chase writes, is intelligence adapted to a specific business, not generic intelligence.
He reaches for a supply chain analogy to make the point concrete. A retailer does not manufacture every truck, ship, or warehouse robot, but it owns the system that determines sourcing, inventory movement, demand forecasting, and customer experience. AI works the same way, in his telling. Companies can buy models, compute, and infrastructure, but they need to own the system that turns those inputs into intelligence specific to their business.
The Three Layers of an Agent System
Chase breaks ownership into three parts: controlling the agent system, owning economics, quality, and risk, and compounding intelligence over time. He defines an agent as the bridge between raw intelligence and work. The model provides intelligence, the harness determines how that intelligence is applied, and context tells the system what matters in a given situation.
The first layer is the model. Owning it can mean using open-weight models when sovereignty, portability, or deployment control matters. It can also mean preserving model optionality, the ability to switch providers as quality, cost, latency, and privacy requirements change. Chase calls optionality defensive because it avoids lock-in, and offensive because it lets a company adopt the best model as soon as it appears.
The second layer is the harness, the orchestration logic that turns model outputs into actions. The harness controls how context is used, including routing, tool use, workflow steps, skills, and more. Much company-specific behavior lives in the harness, Chase writes. If the harness is closed, you are accepting someone else's assumptions about how your agent should work.
The third layer is context: documents, policies, tools, skills, user preferences, organization-level knowledge, and memory. Memory matters because it makes the system more useful over time. If you do not own context and memory, Chase writes, you do not own the intelligence your system accumulates. To own the agent system, he concludes, a company needs control over all three layers: model, harness, and context.
Cost, Quality, Boundaries, and Observability
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Once AI does real work, Chase argues, it has to be managed like any operating system. That means controlling cost, measuring quality, defining where the system can act, and knowing what happened.
Cost matters because intelligence is only valuable if it is cheap relative to the return it generates. AI adoption has increased rapidly, and so have costs. Chase cites Uber as a cautionary tale on that front. Being able to lock down costs at the user, org, or agent level is crucial to scaling AI reliably, he writes.
Quality needs to be measured, not assumed. When a company upgrades a model, changes a prompt, adds a tool, or adjusts a workflow, it needs to know whether the system improved or regressed. If you are unable to measure this, Chase writes, you are unable to improve your agent systematically, which means you are unable to truly control and own it.
Boundaries define where AI can act independently versus where it needs supervision. Companies need to control what data agents access, what tools they use, which actions require approval, and when to escalate. Observability makes the system accountable. If an agent takes an action, the business needs to see what it saw, what it did, what tools it called, and why. Chase ties observability to improvement, audit, and trust.
The Loop That Compounds
The final piece of ownership is the learning loop. The best AI systems get better with use, Chase writes, and the 100th interaction should be more valuable than the 1st because the system has learned about users, workflows, policies, failures, and preferences. Companies need to own both what the system has learned and the learning process itself.
That loop starts with traces. Traces show what an agent did: the context it saw, the tools it called, where it got stuck, and what it produced. Feedback gives those traces meaning, marking them as useful, accepted, rejected, inefficient, risky, or wrong. Together, traces and feedback become the raw material for improvement. They can change prompts, the harness, and tools. For every change, Chase writes, you should be adding an eval to capture that change and make sure future changes do not cause regressions.
At a bare minimum, these learnings need to be owned by the company and portable from model to model and system to system. Even more, Chase writes, you should own this entire loop yourself. It does not really matter how well the agent does the first time someone uses it, in his framing. It matters how much better it is the 100th time they use it.
A Ten-Question Checklist
The essay closes its argument with a checklist for companies assessing their own position. Chase asks whether a company could easily switch to a brand new model provider's state-of-the-art model tomorrow. He asks whether, if a model provider deprecates a model, the company could host it itself to avoid disruption. He asks whether the company knows and controls how its orchestration logic works, such that it knows and controls exactly what goes into the LLM at every step.
Further questions cover portability across clouds and devices, per-user monitoring and control of AI spend, and whether a full trace of all steps and internal logic can be shown when someone asks why an agent did something. He asks whether evals are in place to be confident that swapping models will not introduce regressions, and whether a user's 100th session with an agent will be a better experience than the first because the system knows more about them. The last two questions concern portability of learnings to a different system entirely and control over how the agent learns.
The strategic conclusion follows from the checklist. Companies do not need to build every layer of the AI stack. They should buy the parts that are hard, generic, and undifferentiated. They need to own the layer where advantage compounds: the agent system, the governance around the work it does, and the feedback loop that makes it better with use. Any company can adopt AI, Chase writes, but the companies that create lasting value with AI will be the ones that own their intelligence.