Build the Agent or Power the Agent?

2024-09-30 · Source: Tanay’s Newsletter · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Entrepreneurship & Start-ups, Project & Product Management · Depth: Intermediate, medium

Summary

The article explores a core strategic decision for product builders in the era of horizontal AI agents like Claude Code/Cowork, Codex, and Copilot: whether to "build the agent" or "power the agent." Building the agent means creating a standalone core interface, suitable when a product is already a system of record or offers highly specialized domain reasoning, exemplified by vertical players like Harvey for lawyers. This approach offers interface ownership but increases adoption friction. Conversely, "powering the agent" involves making a horizontal agent better at a specific domain, typically via MCP servers, CLIs, or APIs, by exposing valuable data/context (e.g., Granola) or capabilities/actions (e.g., Higgsfield). Incumbents like Salesforce, with its April 2026 "Headless 360" announcement, and HubSpot are also adopting this. Ultimately, many companies will likely do both, offering an embedded agent for power users within their product and a headless MCP layer for broader audiences accessing slices of functionality through existing horizontal agents.

Key takeaway

For AI Product Managers or startup founders developing new solutions, your strategic choice between building a dedicated agent or powering existing horizontal agents is critical. If your users primarily operate within platforms like ChatGPT or Claude, focus on providing headless APIs or MCP servers to integrate your unique data and capabilities. Conversely, if your product is a system of record for a specialized workflow, building an embedded agent for your power users can solidify your interface ownership and value proposition. Consider a hybrid approach to expand your audience.

Key insights

Product builders must choose between becoming a core AI agent or enhancing existing horizontal agents.

Principles

• Verticalized domain reasoning justifies building a dedicated agent.
• Data/context and capabilities drive value for powering agents.
• Customer workflow location dictates agent strategy.

Method

Founders should assess customer workflow, data ownership, value source (reasoning vs. data/actions), and app-internal work scope to decide on an agent strategy.

In practice

• Expose product data via MCP servers or APIs for horizontal agents.
• Develop embedded agents for power users within your core product.

Topics

• AI Agents
• Product Strategy
• API Economy
• Horizontal Agents
• Vertical AI
• MCP Servers

Best for: CTO, VP of Engineering/Data, Executive, Entrepreneur, Director of AI/ML, AI Product Manager

Related on AIssential

• AI Startups vs. Big Chatbots — With Olivia Moore — AI is reinventing the entire technology industry, offering diverse opportunities for specialized startups despite big chatbot dominance.
• The Builder-PM Manifesto — Product management is fundamentally changing as AI's "harness" operating layer shifts scaling laws from models to team-level strategic framing.
• The Best Way to Talk to Your AI Agents — The shift to AI agents redefines knowledge work, emphasizing staging conditions over direct output production.
• Building AI Agents: A Journey with Sam Bhagwat — AI agent development requires understanding core concepts, organizational design, and deep integration into customer processes for success.
• Building an AI Startup Is Easy. Selling It Is Hard — When product building cost is near zero, distribution, not the product, becomes the critical competitive moat.
• The Agent Era: Building Software Beyond Chat with Box CEO Aaron Levie — Software must fundamentally adapt for an agent-first world, prioritizing robust backend systems over human-centric interfaces.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Tanay’s Newsletter.