Age of Agent Wars

Consumer agents will fight over demand. Enterprise agents will fight over context

2025 was the year of agents. Claude Code and Codex have become very reliable and powerful. Many companies have released their MCP servers, open APIs, and their own agentic interface. This creates multiple access points for the same task: a user can ask a general-purpose agent to use a company’s APIs and complete the job, or they can use the company’s own agent. That is where the agent wars begin.

In consumer markets, this is a fight over demand and discovery. Shopify is a good example. Around three-fourths of its revenue comes from Merchant Solutions, which includes some discovery products like App Store ads and Shop Campaigns.Shopify is also trying to become infrastructure for agentic commerce through UCP.. If product discovery moves from merchant websites, Shop, Google, or Instagram into a general-purpose agent, Shopify may still win through checkout, payments, and merchant infrastructure(which is their core). But it may lose some control over product ranking, sponsored placement, and discovery monetization. The question becomes: if another agent brings the customer, who owns the discovery revenue?

Swiggy and Zomato face the same issue in food delivery. MCP is great for the user: “order me a healthy dinner under ₹400” is easier than opening the app, browsing restaurants, comparing offers, and building a cart. But for the app, that interface is valuable. It controls restaurant ranking, offers, ads, cross-sell, payments, and habit formation. So MCP enabled thrid party agents may increase convenience and orders, but it also risks reducing the surface area where marketplaces monetize attention.

Airbnb shows the other side. Not every agentic experience should be a text box. Travel is high-consideration: users need maps, photos, dates, prices, trade-offs, group planning, and comparison. Brian Chesky has argued that chat is not the right interface for travel and e-commerce; these categories need richer, more visual, interactive AI experiences. For low-consideration commerce, chat may be enough. For high-consideration decisions, the company’s own visual agent may win.

Enterprise is a different battlefield. General-purpose agents are powerful, but they often lack the depth, reliability, and consistency required inside a company. Every enterprise has its own workflows, permissions, internal language, and unwritten operating rules.

This is why company-specific agents often work better. Sentry’s David Cramer has made a similar point: an internal agent, with the right instructions and context, can outperform a generic coding agent for company-specific workflows.

We have seen the same thing internally at our startup. We have deployed our own agent flow, even though Codex / Agents SDK runs in the background. The difference is that our agent knows what to do and where to look: Sentry, CloudWatch, the database, and GitHub issues. It is not just a smarter model; it is a more defined workflow.

So the right answer is not simply “use Codex” or “build everything yourself.”, it is to build for the use case. Use the best agentic stack available, but wrap it with the right context, tools, instructions, permissions, and evaluation.

In consumer markets, agents fight over demand. In an enterprise, agents fight over context. And that is where the real agent wars begin.