In a recent conversation with Kwindla Hultman Kramer, CEO of Daily, we explored the evolving landscape of voice AI infrastructure and the critical decisions companies face when choosing their tech stack. Daily, a developer tools platform for real-time voice, video, and AI, has been at the forefront of voice technology since 2016. They're also the team behind Pipecat, the widely-used open-source framework for voice customer service agents.

The Evolution of Voice AI Infrastructure

The voice AI landscape has undergone rapid transformation, particularly since the release of GPT-4. As Kwindla explains, Daily recognized early on that the industry needed robust open-source tooling to handle the complexities of voice AI systems. These challenges include phrase endpoint detection, interruption handling, low-latency network transport, and pipeline observability.

This recognition led to the creation of Pipecat, which has become a cornerstone of the voice AI ecosystem. Built with Daily's extensive experience in media processing pipelines, Pipecat offers state-of-the-art implementations of low-level building blocks, all wrapped in an extensible and flexible framework.

Making the Build vs. Buy Decision

When it comes to choosing between building in-house, using Pipecat, or opting for a hosted solution, Kwindla offers clear guidance: "Build your product, not things that are not core to your product."

For companies just starting their voice AI journey, Kwindla often recommends using full-stack platforms like Vapi, which offer a low-code approach and quick time-to-market. However, as companies mature and require more granular control, they might consider transitioning to Pipecat or building custom solutions.

The key factors that typically push companies toward building their own solutions include:

1. Regulatory and compliance requirements, especially for larger enterprises

2. Integration needs with complex backend systems or large internal knowledge bases

3. Control over unique IP and model-level specifics, particularly when working with custom-trained or fine-tuned models

The Case Against Building from Scratch

In 2025, building low-level voice infrastructure from scratch rarely makes sense. As Kwindla notes, "2024 was all of us trying to figure out the right patterns and specific algorithms. In 2025, we've all popped up a level and are now thinking more about things like the long tail of agent performance and integration with new function calling capabilities."

The open-source community has already invested significant engineering hours into solving these foundational challenges. Companies starting today would be better served focusing their resources on their unique value proposition rather than reinventing the wheel.

The Importance of Observability

One crucial aspect often overlooked in voice AI systems is observability. Without proper monitoring tools, it's challenging to understand how your product performs in real-world conditions. As Daily's experience shows, companies often underestimate the impact of performance issues on their bottom line.

"The patterns are always different than your engineers and QA team," Kwindla emphasizes. "Figuring out how to build those kinds of last mile tools is so important."

Introducing Pipecat Cloud

Recognizing the challenges companies face in deploying and scaling Pipecat agents, Daily has introduced Pipecat Cloud. This hosting platform allows companies to maintain the flexibility of Pipecat while offloading the DevOps complexity of running distributed systems at scale.

Pipecat Cloud targets a specific niche: companies that want to build their own agents but prefer not to handle the operational complexity of global deployment and scaling. The solution maintains Pipecat's commitment to flexibility – all code remains portable and can be run on any infrastructure.

Looking Ahead

As voice AI continues to evolve, the infrastructure stack is maturing. Companies now have more options than ever, from full-stack platforms to open-source frameworks and cloud-hosted solutions. The key is choosing the right tool for your specific needs while maintaining the flexibility to adapt as those needs change.

Whether you're just starting your voice AI journey or looking to scale existing solutions, the ecosystem now offers mature, battle-tested options that can help you focus on what matters most – building great products that deliver value to your users.