Follow me on LinkedIn - AI, GA4, BigQuery

Conductor is the built-in AI assistant from Retell that helps you build and test your Voice Agents using simple English.

Conductor is free to use up to 30 messages per day. But after that, it is $2 per 10 messages, making it the most expensive AI assistant in the market.

If you send 50 Conductor messages per day, you’d pay about $120/month on top of your normal Retell usage. 

If you send 100 Conductor messages per day, you’d pay about $420/month on top of your normal Retell usage.

So if you heavily use Conductor, it can cost you a lot more than the most expensive Claude Max plan.


There is also a 200 messages per workspace per day cap.

Note: If pay-as-you-go is turned off, Conductor stops answering after the free allowance rather than charging overage.

The issues you can run into while using Retell AI Conductor.

#1 Underspecifying your requirements.

If you under-specify your requirements to Retell Conductor, the voice agent you build will not work reliably in production. 

Example:

Underspecified:

“Build me a receptionist that handles calls.”

A good starting point:

“Build me a receptionist for my dental practice that answers both during business hours and after hours, asks the caller’s name and reason for calling, routes sales calls to the sales team, routes booking calls to the booking queue, and transfers urgent calls to a human immediately.”

  • You must clearly define what the voice agent must do.
  • Clearly define what the agent must not do (e.g. legal advice, medical advice, billing decisions), which in most cases is far more important than what the agent must do.
  • You must have documented mandatory steps, confirmations, and escalation rules in advance even before you start building the agent. Otherwise your agent would fail in production.
  • You must anticipate how callers will mispronounce words, go off-script, or mix multiple intents, so your agent does not fail during messy real calls.

This is why many “quickly built” agents sound fine in test calls but break badly in production, hallucinating policies, misclassifying urgency, or mishandling compliance-sensitive conversations.


Lack of expertise in voice AI development often results in under specification.

Without domain expertise, self-built voice agents are much more likely to be unreliable, brittle, and to hallucinate or mishandle real-world edge cases, especially in regulated or operationally complex domains.

Retell’s own Conductor docs explicitly say the quality of the output depends heavily on the precision of the initial requirements.


Why does under-specification hurt?

  • Conductor makes assumptions when the brief is vague. That can be fine in a demo, but in production those assumptions often become wrong branch logic, weak handoff rules, or missing edge-case handling.

  • Voice AI agents need explicit flow design. Retell’s docs recommend breaking conditions into simpler pieces and using transition examples or global nodes when the flow gets complex.

  • Long or fluffy prompts reduce reliability. Retell’s own best-practice docs say concise prompts work better, and if instructions are too complex, you may need a more capable model or better structure.

#2 Increase production and testing cost.

Underspecifying your requirements can significantly increase both your testing and production costs when using Conductor, because you’ll spend more time iterating, correcting assumptions, and retesting than an experienced voice AI developer would. 

Any money saved by avoiding a developer can quickly get lost in DIY trial and error, paid testing, and production usage.

Even after all that extra testing and iteration, the agent can still be brittle in production if the build lacks domain expertise. 

Without a deep understanding of the business, call flows, edge cases, and caller behaviour, the agent may work in demos but fail when real-world complexity shows up.

Areas where the Retell AI Conductor excels.

If you know what you are doing, Retell Conductor can be a great tool for testing voice agents. 

Once you ask Conductor to test an agent, it will propose test cases for you to review and approve. 


Once you approve the test cases, it will first write the test cases, then review test case definitions, run the tests, wait for the tests to finish and check test results.

Then, based on the test results, it will review the agent global prompt. 

Then it may or may not recommend changes.


You can attach a real past call to Conductor, and it can see what actually happened and suggest improvements. This is one of its greatest strengths.

Conductor can automatically convert a Single Prompt agent into a Conversational flow agent.

You can ask Conductor to automatically convert a single prompt agent into a conversational flow agent: