
Prompting Text-to-Speech Models in 2026
Amongst other innovations, new text-to-speech models have introduced new capabilities and controls. In voice agents, this means leaning into new prompting best practices and patterns.
Josh Lawman

AI Engineer
Josh Lawman is the founder of Adder, a studio dedicated to solving business problems through applied AI. An engineer with 8 years of experience building ML & AI products. Having played with LLMs since the GPT-3 era, Josh combines a passion for AI systems with analytical rigor. He helps companies build AI products that are secure, measurable, and ready to scale.

Amongst other innovations, new text-to-speech models have introduced new capabilities and controls. In voice agents, this means leaning into new prompting best practices and patterns.

We propose RandomMCP to increase sub agent randomness and diversity, thereby improving the ability of developers and systems to leverage larger numbers of sub agents effectively.