More AI Agents Does Not Mean Better Results — Here Is What the Numbers Actually Say

More AI Agents Does Not Mean Better Results — Here Is What the Numbers Actually Say

Imagine paying $20,000 for a task that one AI agent could have completed for a fraction of that cost. That is not a hypothetical horror story. It actually happened, and the person who called it out publicly works at OpenAI.

Eric Provencher, a developer on OpenAI's Codex team, issued a pointed warning in September 2026 about what he calls the "coordination tax" in agentic AI workflows. His argument is direct: running more than two parallel sub-agents almost always burns tokens without improving the quality of the output. The reason is a trust problem. Agents working in parallel do not trust each other's work, so they end up re-verifying everything the other agents have done. As Provencher put it, they "double-check everyone's homework." The result is massive token consumption with no measurable gain in quality.

The most striking example Provencher cited involves a project where 1,393 Fable agents were deployed to refactor a single Python file, burning through $20,000 in tokens to complete it. Provencher noted that a single Astra agent could have handled the same job for a fraction of that cost. He also flagged that system prompts compound the problem at scale because they add token overhead across every sub-agent, and without sufficient context, those sub-agents make redundant tool calls on top of each other. His suggested fix is a cleaner architecture: delegate tasks to separate threads and have them notify the main agent only when finished, rather than constantly polling for status updates. He acknowledged that OpenAI still needs to ship better solutions to address this structurally.

For small and mid-size business owners experimenting with AI-powered workflows, this is a genuinely important signal. The allure of "more agents equals more power" is easy to buy into, especially when vendors and platforms market multi-agent systems as a premium capability. But Provencher's warning reveals that scaling up agents without a deliberate architecture is not just inefficient. It is a cost trap that can multiply your AI spending dramatically while delivering the same result you could have gotten from a single, well-scoped agent.

This matters directly for your marketing operations. If your team is using AI tools that run parallel agents to handle content creation, lead qualification, or campaign analysis, you may be paying token costs far above what the actual task requires. The insight here is not that AI agents are bad. It is that agent architecture needs to be intentional. Fewer, better-contexted agents with clear task boundaries consistently outperform sprawling agent networks that spend most of their capacity verifying each other's work.

The strategic takeaway for business owners is that efficiency in AI is not about volume. It is about precision. A well-designed single-agent workflow with a strong system prompt and clear task scope can outperform a swarm of loosely coordinated agents at a fraction of the token cost. Before you upgrade to a larger or more complex multi-agent setup, ask your vendor or implementation team exactly how many parallel agents are running, what each one is doing, and whether the architecture has been tested for token efficiency versus a simpler single-agent alternative.

This week, review one AI workflow you are currently running or considering. Ask specifically how many agents or parallel processes are involved and what the estimated token cost per task is. If you cannot get a clear answer, that is itself a red flag worth addressing before you scale.

Getting AI right for your business is not about using the most sophisticated setup on the market. It is about using the right setup for your actual goals, at a cost that makes the investment worthwhile.

Originally inspired by: AI agent swarms are a massive waste of tokens with zero quality gain, says OpenAI Codex developer (https://the-decoder.com/ai-agent-swarms-are-a-massive-waste-of-tokens-with-zero-quality-gain-says-openai-codex-developer/) See how Leads to Conversion can help you build smarter, leaner AI workflows that actually grow your business. Schedule an a chat

Your turn

What is your traffic actually doing?

Send us your details and we will come back with a short, specific read on what your traffic, your pages and your pipeline are doing today — and the first three things we would change. A real person reads every submission, and you get the read whether or not we ever work together.