Hamming
Request a demo

Maven AGI ships enterprise voice agents with Hamming

Maven AGI builds enterprise-ready AI agents for the full customer journey, focused on complex, high-friction enterprise environments.
Maven AGI
Location
Boston, Massachusetts
Industry
Enterprise AI
Funding
Series B ($50M, $78M total)
Website
mavenagi.com
Use cases
  • End-to-end voice agent testing
  • Concurrent load testing
  • Background noise validation
  • CI/CD release gating
“It's not just time saving, it's the capability and velocity. You can't even do it without a platform like Hamming.”
Dongsheng Wang
Tech Lead, Voice AI, Maven AGI

About Maven AGI

Founded in 2023 by executives from HubSpot, Google, and Stripe, Maven AGI has raised $78M from investors including Dell Technologies Capital, Cisco Investments, and Lux Capital. It serves leading enterprises, including Tripadvisor, Rho, and Clio.

Hamming became the testing platform that helped Maven AGI scale its quality bar. The team can now:

  • Run hundreds of concurrent test calls
  • Catch regressions early with CI/CD smoke tests before shipping
  • Validate agent quality with built-in background-noise simulation and other voice-specific tests

The challenge: maintaining a simulation and testing platform is not easy

Maven AGI has treated voice quality and reliability as core product requirements from day one. Long before scaling the platform, the team invested heavily in automated simulation and testing, because it saw early that great voice experiences depend on rigorous validation under real-world conditions.

The voice team built a homegrown simulation and smoke-testing system alongside the first version of the voice agent. With a small engineering team, they moved quickly and even used manual calling as a load test, asking tens of team members to place calls. That was useful, but far from true concurrency or production-scale testing.

As the product matured, the testing roadmap split into two demanding categories. Voice-specific requirements, such as more language support, background-noise simulation, and large-scale concurrent call testing, needed specialized infrastructure. Table-stakes platform work, such as repeatable test management, automated regression testing, and CI/CD integration, carried its own build-and-maintain cost. Rather than pull engineering focus from its voice agents and customer products, Maven AGI chose a purpose-built testing platform that provided both out of the box.

“We were testing for a customer under high pressure for background noise. Then we saw a Hamming background noise feature we didn't notice before. Maybe it was released last week. We used it, and it was awesome.”
Dongsheng Wang
Tech Lead, Voice AI, Maven AGI

Before and after Hamming

MetricBefore HammingWith Hamming
Testing methodBefore HammingHomegrown automated simulation system and manual testingWith HammingAutomated parallel test runs on demand
Load testingBefore HammingManual load test: tens of team members placing callsWith Hamming100 true concurrent calls in a single session
Background noise testingBefore HammingManual: playing a video in the background during a live callWith HammingOn-demand simulated background noise
Release and regression checksBefore HammingHomegrown automated simulation systemWith HammingCI/CD smoke tests flag issues, and the team investigates before shipping

How Hamming helps Maven AGI

  • 01

    Automated end-to-end testing for complex workflows

    Maven AGI's agents handle multi-step enterprise workflows, seven or eight steps deep, where each step must execute correctly for the conversation to succeed.

    Hamming handles the testing layer, so the team can add a suite of complex tests quickly.

    Guide to AI voice agent quality assurance

  • 02

    Concurrent load testing at scale

    Hamming gave Maven AGI true concurrent load testing on demand. The team routinely runs hundreds of parallel calls in a single session to stress-test its infrastructure.

    When a customer asked Maven AGI to validate 100 concurrent calls, the team delivered, turning a difficult customer request into a routine one.

    Complete voice agent testing guide

  • 03

    Background noise and environment simulation

    A customer raised a background-noise edge case that had to be reproduced, fixed, and validated under pressure. Hamming's simulated noise environments let the team run those scenarios on demand and ship the fix with confidence.

    The team now tests against simulated noise on demand, a category of testing that wasn't practical to cover before.

    Background noise testing for voice agents

  • 04

    CI/CD integration

    Maven AGI added Hamming to its CI/CD pipeline to run smoke tests with every release. If the tests aren't green, the team investigates before deploying. The flow isn't fully automated yet, but it gives the team an early signal to catch problems before they ship.

    Beyond CI/CD, the team uses Hamming's UI for deeper exploratory testing when tuning agents for specific customers. Automated smoke tests plus on-demand testing give the team confidence to move faster.

    Voice agent regression testing

Why quality matters to Maven AGI

Customer expectations for voice agents are rising fast. A year ago, a realistic conversational voice agent felt groundbreaking. Today that level of quality is no longer impressive; it is the minimum users expect.

That shift is why Maven AGI invested early in systematic testing and evaluation. The team keeps a high quality bar and refuses to ship voice experiences that feel unreliable, unnatural, or unfinished. As expectations rise, catching regressions, validating performance across environments, and stress-testing under load have become essential parts of development.

“Yesterday was amazing. Today it's just expected.”
Dongsheng Wang
Tech Lead, Voice AI, Maven AGI

Why Maven AGI chose Hamming

Maven AGI evaluated five or six platforms and ran real test scenarios against the top two. Hamming won on product polish, data model, and how well it matched the team's mental model of agent testing.

The build-versus-buy decision was clear. As Dongsheng put it: “We're builders, we want to build stuff. But when we seriously looked at this as a startup, buying made more sense.”

  • The data model (agent to test cases to results to monitoring) matched how the team thinks about voice agent quality
  • A more polished product and UX than the alternatives evaluated
  • Works for developers, solution engineers, and PMs, not just one persona

The results

  • 01

    An FTE reclaimed, plus infrastructure work avoided

    Manual testing alone consumes 0.5 to 1 FTE. That excludes the engineering time to build, operate, and keep expanding production-grade voice testing infrastructure, from concurrent-call orchestration and background-noise environments to regression tooling and CI/CD integrations. With Hamming, Maven avoids both costs: day-to-day manual validation and an ongoing internal testing-platform investment.

  • 02

    Hundreds of concurrent calls on demand

    Validating 100 concurrent calls took a single session, and the team routinely runs well beyond that to stress-test its infrastructure. Manual testing couldn't get close.

  • 03

    Voice-specific testing, ready when needed

    Simulated noise environments, language coverage, and concurrent-call testing let Maven validate real customer conditions without building another specialized testing system.

What's next

  • 01

    A QA agent that calls Hamming on its own

    Maven AGI is building an internal QA agent to verify changes automatically. Rather than hardcoding which tests to run, the agent will decide what to validate based on the change and call Hamming's MCP endpoints to run the right tests.

  • 02

    Closing the guardrails gap

    Maven AGI's testing today focuses on making sure voice agents reliably complete intended tasks and that new releases don't introduce regressions in quality or behavior. As the platform matures, the team is expanding coverage of guardrails for customer workflows and edge-case interactions, including more sophisticated adversarial prompts and jailbreak attempts.

    This work builds on Maven AGI's existing detection and prevention measures. Working with Hamming helped identify ways to make systematic red-teaming and adversarial evaluation a stronger part of the broader safety and reliability roadmap.

“We get new features so quickly, you guys are building things on the fly. All the features we ask for, they turn around in one or two weeks. Looking back, we made the right decision. We're really happy.”
Dongsheng Wang
Tech Lead, Voice AI, Maven AGI