Business Development Representative

Phonely's Alma: A voice AI model that can handle messy phone calls. Sep 1, 2026 Phonely introduced Alma, a voice-specific large language model trained on over 10 million phone calls. Designed to handle background noise, interruptions, and awkward conversational shifts, the company claims Alma delivers significantly lower latency and reduced costs compared to general text models. Voice AI startup Phonely is taking on the industry giants with a new language model built strictly for phone calls. Called Alma, the system steps away from traditional text-focused LLMs to focus entirely on automated phone interactions, promising faster responses and much lower operating costs than traditional setups. According to Phonely, most voice agents today still run on general-purpose text models that struggle with real-world conversations. While frontier labs have poured resources into coding and reasoning tools, phone agents still get tripped up by everyday call habits - like someone talking over the assistant, changing their mind mid-sentence, or dealing with background noise. Ditching text models for real phone calls. To fix those awkward pauses, Phonely trained Phonely Alma voice LLM OpenAI competitor on a dataset of over 10 million real phone conversations rather than clean text articles. CEO and co-founder Will Bodewes explained that because existing market models were built for other tasks, the team decided to build a custom model trained directly on real call traffic. The performance stats highlight a clear focus on speed and cost. Traditional text models usually take around 600 milliseconds to output their first token, creating an unnatural delay. Alma cuts that initial response time down to between 50 and 150 milliseconds. This helps to keep the back-and-forth flow natural. Phonely also claims the model is 61% faster and over 80% cheaper than OpenAI's latest offerings. The firm notes that one client saw sales conversions rise 3% after switching, driving an extra $300,000 in monthly revenue. Self-improving agents for high-stakes calls. Beyond raw speed, the platform is designed to learn from live interactions. Phonely runs the model across millions of active conversations. So, the system identifies where calls break down and feeds those lessons back into the pipeline. That allows agents to self-improve on live traffic without waiting for a frontier lab to ship its next base model. The feature set arrives at a crucial time for industries relying on phone transactions. This includes legal intake, insurance, appointment bookings, and sales. With roughly $2.7 billion transacted over the phone daily across those sectors, reducing missed rebuttals and lag is a huge financial priority. To keep performance steady during heavy usage, Alma includes built-in failovers, multi-region routing and priority inference. It works alongside any existing transcriber or text-to-speech software, removing the need for heavy prompt engineering. While Alma already powers 100% of Phonely's internal traffic, enterprise teams can now get custom deployments with dedicated rate limits for their own setups.

Coval launches the first self-improving agents, powered by phonely's new Alma model. Coval Team Coval and Phonely are giving voice agents a continuous feedback loop, closing the gap between production performance and voice agent improvement. New collaboration gives voice agents a continuous feedback loop to improve customer interactions in real time SAN FRANCISCO, September 1, 2026 - Coval, the evaluation platform for voice AI, today announced the first-ever self-improving voice agents, enabling them to continuously identify and fix performance gaps based on real-world conversations. As the voice AI agents market grows from $2.4 billion in 2024 to a projected $47.5 billion by 2034, enterprises are moving voice AI into increasingly critical customer interactions, making continuous improvement essential. Coval is bringing this capability to market through a new collaboration with Phonely, combining Coval's continuous evaluation platform with Phonely's voice models to create the first feedback loop between production performance and voice agent improvement. The collaboration connects Coval's evaluation platform with Phonely's new voice model, Alma, a large language model purpose-built for voice agents. Together, these technologies give businesses, for the first time, a way to evaluate how their agents perform in real customer conversations and use those insights to improve agent performance over time. Alma is 61% faster than GPT-5.2, more than 80% cheaper than OpenAI's GPT-5 models, and trained on more than 10 million phone conversations to better understand how people communicate in real-world voice interactions. Coval evaluates real production conversations, identifying where an agent is performing well and where it's falling short. The findings are then fed back into Alma to retrain the agent, allowing it to learn from recurring patterns and automatically improve without requiring teams to manually identify issues or continuously reprompt the agent. "We initially met the Phonely team when we were in the same Y Combinator batch, and since then, our companies have grown up together," said Brooke Hopkins, Founder and CEO of Coval. "Phonely was also one of Coval's earliest customers, so we've seen firsthand how important it is for voice agents to constantly be improving and adjusting their interactions based on customer needs. Bringing Coval and Phonely together to close that loop and create a better customer experience is something we've been working toward for a long time, so it's incredibly exciting to see it come to life." "The best voice agents are the ones that can self-improve over time by learning from the conversations they have with real customers. With Alma, we're giving businesses a model built specifically for voice, and Coval gives them a way to see what's working, what isn't and where the agent can improve," said Will Bodewes, Co-Founder and CEO of Phonely. "Brooke and I have been building alongside each other since we met, so I'm especially grateful that we get to work together on something we've both cared about from the beginning." From voice agent measurement to process improvement. As businesses transition customer interactions to voice AI, the ability to understand how agents perform in real conversations and continuously improve them is becoming a crucial part of building reliable customer experiences. A new Gartner survey finds that customer service leaders see GenAI voicebots becoming one of the most valuable channels over the next two years, behind only GenAI chatbots and live chat. At the same time, companies are putting significantly more spend behind AI, with customer service organizations increasing AI spending by 38%. Today, most voice agents rely on general-purpose foundation models, such as OpenAI or Gemini, to generate responses in the middle of a conversation. Those models are powerful, but the agents themselves remain largely static and are not trained on the customer's specific use case. A business can evaluate its conversations and identify recurring failures, but turning those insights into continuous model-level improvement is still a significant challenge. Now, Phonely voice agents can use Alma to generate responses during live conversations instead of a general-purpose model. Coval then evaluates those production conversations to identify what's working, what's not, and where the agent has opportunities to improve. Those findings can then inform ongoing updates to the customer's model, helping the agent adapt to the specific behaviors and patterns of the business. Customer data is used exclusively to improve that customer's own model and is never used to train or improve other customers' agents. The process moves from: * Build | deploy | evaluate | manually fix | repeat * Build | deploy | evaluate | auto-improve | validate | redeploy | repeat What comes next for voice AI. The shift from static voice agents to continuously improving agents represents a broader change in how voice AI will be developed and managed. The next generation of voice agents will become more capable over time as businesses learn from real customer interactions and use those insights to improve their performance. This is the direction Coval is building toward, with an improvement process in which production conversations provide both the data and the evidence needed to understand what should improve next. With Phonely and Alma, Coval is taking another step toward a future where voice agents can learn from how they are actually being used and become more effective and reliable over time. About Coval Coval is the evaluation platform for voice AI quality, helping enterprises simulate, evaluate, monitor and improve their AI voice agents at production scale. Founded by Brooke Hopkins, who previously led evaluation job infrastructure at Waymo, Coval brings the simulation-first discipline that made autonomous vehicles safe for public roads to the next generation of autonomous AI agents. Founded in 2024 and a Y Combinator graduate, the company is trusted by over 60 organizations, including Zoom, Deepgram and Fortune 500 enterprises, to ensure their voice agents work as expected before and after deployment. Coval is based in San Francisco and has raised a total of $31 million from Norwest, Base10 Partners, Twilio Ventures, Swift Ventures, and Y Combinator. For more information, visit www.coval.ai. About Phonely Phonely is an AI-powered voice agent platform based in San Francisco. The company helps businesses answer every call with intelligent, natural-sounding AI agents that book appointments, resolve inquiries, and drive revenue. Learn more at phonely.ai.