The AI industry just broke into four camps over the most important question nobody is talking about: What should it cost to rent an AI worker for an hour?

Anthropic announced it is charging $0.08 per session hour for Claude agents. OpenAI responded by open-sourcing its agent harness framework. Google and Microsoft are bundling agents into enterprise packages with no per-hour fees. Meta has not announced pricing at all.

These are not small differences. They represent four fundamentally incompatible visions of how AI work gets priced, sold, and controlled. And the stakes could not be higher.

The Economics Problem Nobody Saw Coming

AI agents do not work like chatbots. A chatbot answers your question and stops. An agent takes your task and runs with it, potentially for hours, calling APIs, browsing websites, writing code, debugging errors, and iterating until the job is done.

That breaks every existing AI pricing model. Token pricing makes no sense when an agent might use 50,000 tokens or 5 million depending on how complex the task gets. Flat-rate subscriptions get obliterated when power users run agents 24/7.

Anthropic was the first to crack. The company that once promised no ads and transparent pricing just implemented session-hour billing for agent frameworks. Eight cents per hour sounds cheap until you realize a sophisticated coding agent can run for 20 hours on a single complex task. That is $1.60 per task, or $400+ per month for a developer using agents daily.

OpenAI is Playing a Different Game

While Anthropic charges by the hour, OpenAI open-sourced its entire agent harness framework. No licensing fees. No session limits. Anyone can download it, modify it, and run it on their own infrastructure.

This is not altruism. OpenAI makes money when people use GPT models, regardless of whether they pay for the harness. By giving away the orchestration layer, OpenAI encourages heavier model usage while avoiding the pricing complexity that is killing Anthropic's margins.

Google and Microsoft are taking a third approach: enterprise bundling. Instead of charging per hour or giving away software, they are rolling agents into existing cloud and productivity packages. Pay for Office 365 or Google Workspace, get agents included.

The Real Battle is For Control

Pricing strategies reveal who wants to control the AI agent ecosystem. Anthropic is charging developers directly, building a recurring revenue stream that could survive model commoditization. OpenAI is subsidizing adoption to drive model usage and maintain market share. Google and Microsoft are using agents to lock enterprises deeper into their platforms.

For developers, this means chaos. Building on Anthropic agents could get expensive fast. Building on OpenAI means vendor lock-in to their models. Building on Google or Microsoft means your startup competes with your infrastructure provider.

Meta has said nothing about agent pricing, which might be the smartest strategy of all. While competitors fight over how to monetize agents, Meta can let Llama models power whatever framework wins.

What Happens Next

The agent pricing war will not be won by whoever charges the least. It will be won by whoever creates the most value while building the strongest moat.

Anthropic bets that session-hour pricing will fund better models and agent capabilities. OpenAI bets that free harnesses will drive model adoption and create switching costs. Google and Microsoft bet that enterprise bundling will make agents feel "free" while driving platform lock-in.

The winner will determine whether AI agents become an expensive luxury tool for big companies or a commodity service that any developer can afford. That decision will shape who benefits from the AI revolution.

Eight cents per hour does not sound like much. But when it determines who gets to build the future, every cent counts.