Oct 17, 2026 · The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch · 5 min read

LangChain Founder on Agent Economics and Infrastructure

Harrison Chase of LangChain analyzes the shifting economics of AI agents, arguing that value accrues to domain-specific harnesses and verification layers rather than generic models. The discussion covers the necessity of neutral infrastructure, the rise of vertical agents, and the critical importance of cost visibility in enterprise AI adoption.

Business Strategy Technology Infrastructure Venture Capital

The takeaways

Value Accrues to Domain Verification. The highest value in the agent stack lies with domain experts who build verifiers and evals, not just the model itself. Startups should prioritize building context and harness layers before attempting model post-training.
Neutral Harnesses Prevent Lock-In. Maintaining a model-agnostic harness is critical for offensive and defensive flexibility. It allows companies to switch to the best-performing model without losing accumulated memory or context.
Inference Demand Outpaces Supply. The market for inference is significantly larger than anticipated, driven by multi-agent systems. Any reliable inference provider will find success due to overwhelming demand for compute.
Cost Visibility Drives Efficiency. Enterprises must implement three-step cost controls: visibility into spend, budgetary safeguards, and harness-level optimization. This approach reduces waste from inefficient agent loops and prompt caching failures.
Vertical Agents Dominate Consumer Apps. Consumer applications will deploy vertical agents that wrap proprietary APIs and data. These agents provide a natural language interface for horizontal agents to interact with specific services efficiently.
Memory Remains Unsolved and Domain-Specific. General-purpose memory solutions lack product-market fit; effective memory systems must be domain-specific. The core challenge is distilling raw interactions into valuable context, a task LLMs still struggle with.

The Shift to Domain-Specific Value

Harrison Chase, founder of LangChain, argues that the AI agent market is moving away from generic model competition toward domain-specific verification and infrastructure. The core challenge in building agents is not intelligence, but reliability; models are non-deterministic black boxes that require robust harnesses to manage cascading errors. Consequently, value accrues to those who can quantify success through domain-specific evals and verifiers, rather than those who simply deploy the latest frontier model.

Infrastructure and the Neutral Harness

Chase emphasizes the strategic necessity of neutral harnesses. By decoupling the agent logic from specific model providers, companies avoid lock-in and retain the ability to switch to superior models as they emerge. This portability is both an offensive advantage, allowing rapid adoption of new capabilities, and a defensive one, protecting against pricing changes or vendor instability. The context and memory built within these harnesses are the true assets, making the harness layer more valuable than the underlying model for many vertical applications.

Economic Realities of Inference

The demand for inference has exploded, driven by the complexity of multi-agent systems and coding tasks. Chase notes that while model costs are decreasing, total spend is rising due to volume. Enterprises are adopting a three-step cost control framework: visibility into agent spend, strict budgetary controls, and harness-level optimization such as prompt caching. This shift indicates that AI ROI is becoming a measurable operational metric rather than a speculative guess.

Strategic Implications for Founders

For entrepreneurs, the path to success involves starting at the harness and context layer, leveraging domain expertise to build vertical agents. General-purpose memory and routing are currently unsolved or commoditized, but domain-specific implementations offer high ROI. The market favors speed, strong technical talent, and brand momentum, with early movers like LangChain and Cursor establishing durable advantages through rapid iteration and viral adoption. Investors should focus on teams that can execute quickly in niche verticals where AI penetration is low but domain expertise is high.

Key insights

The primary barrier to agent adoption is non-determinism, not intelligence. Agents require robust harnesses to manage cascading errors from natural language inputs and outputs.

Technical Strategy →

Impact: Companies must invest in observability and eval frameworks to ensure reliability, shifting focus from raw model capability to system robustness.
Value in the AI stack is shifting to domain experts who build verifiers and evals. These components allow for the quantification of agent performance in specific verticals.

Market Dynamics →

Impact: Vertical SaaS companies with deep domain knowledge have a competitive advantage in building high-value agents over horizontal generalists.
Neutral harnesses are essential for maintaining model portability. Locking context and memory to a single model provider creates significant strategic risk.

Infrastructure →

Impact: Enterprises can optimize costs and performance by switching models dynamically without rebuilding their agent architecture or losing data.
Inference demand is structurally higher than predicted due to multi-agent interactions. The market size is large enough to support multiple successful providers.

Market Sizing →

Impact: Investors should view inference infrastructure as a high-growth sector with low risk of market saturation in the near term.
General-purpose memory solutions have failed to find product-market fit. Effective memory systems must be domain-specific to distill valuable context from raw interactions.

Product Strategy →

Impact: Founders should avoid building horizontal memory tools and instead focus on vertical-specific learning agents that understand their niche.

Action items

Implement a three-step cost control framework for AI agents: track spend visibility, set budgetary limits, and optimize harness efficiency via prompt caching.

Impact: This approach can significantly reduce wasted spend on inefficient agent loops and improve overall AI ROI for engineering teams.
Prioritize building domain-specific evals and verifiers before investing in model post-training. Start with context and harness engineering to capture immediate value.

Impact: This strategy offers higher ROI by leveraging existing frontier models while building proprietary domain knowledge that is harder to replicate.
Design agent architectures to be model-agnostic, ensuring that memory and context are stored in a portable format independent of any specific LLM provider.

Impact: This prevents vendor lock-in and allows for rapid adoption of new, better-performing models as they become available in the market.
Focus on vertical agent opportunities in industries with high domain expertise and low current AI penetration, rather than competing in saturated horizontal markets.

Impact: Vertical agents can leverage proprietary data and workflows to create defensible moats that horizontal generalists cannot easily match.
Monitor inference costs and demand trends closely, recognizing that total spend may rise even as per-token costs fall due to increased agent complexity.

Impact: Proactive cost management ensures that scaling AI agents does not lead to unsustainable operational expenses for the business.

Quotes

“The hard part about building agents is they don't behave how you want them to behave.”

“I think the main way our business changes as the model gets better is we basically have to erase big parts of our product.”

“I think the reason we need neutral harnesses is that there is a lot of state that you build up either before you build the agent or as part of the agent running.”