When a vendor brags about 13 agents, run
The agent count is marketing, not capability. What the agents are wired into tells you everything.
The agent count is marketing, not capability. What the agents are wired into tells you everything.
The new pitch in martech is the agent count. A vendor gets you on a call and announces they now have thirteen agents available. A research agent, a copy agent, a segmentation agent, an optimization agent, and nine more you'll never use. The number is supposed to impress you. It should do the opposite.
Thirteen agents is a red flag. It usually means thirteen thin wrappers around a model someone else built, bolted onto a product that needed a story for the AI era. The count is marketing, not capability: a way to look like you're at the frontier without doing the hard part.
Here's our honest position: we don't think anyone can out-build Anthropic, OpenAI, and Google on the actual models. That's a capital-and-research race measured in billions, and it's already being run by people who are very good at it. Pretending a martech company is going to win it is how you get thirteen agents that are really one mediocre prompt thirteen times.
The opportunity isn't building the model. It's the tooling around the model: exposing your product through MCP, a CLI, the interfaces that let a capable AI actually drive your system and let your customers compose it into their own workflows. Some companies are quietly doing this real work. Others are doing what we'd call agent washing: slapping "agentic" on the same old features and counting them up.
Ask what the agents actually do that the product couldn't do before. Ask whether you can drive the thing from your own tooling, or whether you're locked into their chat box. Ask who built the underlying model. And if the answer is "we did," ask how, because that's an extraordinary claim.
A real AI strategy makes your product controllable and composable. Agent washing just makes the deck longer. The number of agents tells you nothing. What they're wired into tells you everything.
Agentic sprawl is the vendor pattern of shipping many named "agents" — a research agent, a copy agent, a segmentation agent — that are mostly thin prompt wrappers around a general-purpose model. The agent count is a marketing number. What matters is what the agents are wired into: whether the product exposes real, composable operations an AI can drive.
Not building a frontier model — that race belongs to the model labs. The durable work is the tooling around the model: exposing the product's real operations through MCP and APIs so a capable AI (and the customer's own agents) can drive the system end to end. Controllable and composable beats a longer list of branded agents.