The evidence.
Everything on this site comes from pages any visitor can open, read on November 22, 2026. I created no account, submitted no form and tested no system.
Where Cresta already leads
- Testing that starts from written requirements.Each requirement is a yes or no check marked must-have or nice-to-have, and each grader gets an accuracy score against human review. AI Agent Testing 2.0, June 4, 2026
- Test callers built from the customer’s own conversations.Synthetic Customers, with test cases generated from production failures and reviewer feedback. Test and deploy
- Voice tested on a bad line.Soundcheck simulates noise, interruptions and poor connections, and the same graders score live calls after launch. Soundcheck, November 19, 2026
- It tells customers how to measure honestly.Its financial services playbook says containment is easy to inflate, and how to count it so that it holds up. Part I, Part II
- It names the people who deploy.Five roles, including this one, and three ways to share ownership with the customer. AI Agent implementation
- Independent checks on the agent itself.ISO/IEC 42001 for how AI is governed, and AIUC-1 for the agent, announced November 4, 2026. Trust, AIUC-1
Cresta beside six rivals
What a customer’s deployment lead can establish from each vendor’s public pages before signing. Each name links to one of that vendor’s own pages.
| Vendor | Tests before launch | Voice conditions | Release controls | After launch | Public docs | Assurance stated | Who deploys | Pricing |
|---|---|---|---|---|---|---|---|---|
| Cresta | Automated testing with Synthetic Customers and graders scored against human review | Soundcheck: noise, interruptions, poor connections | Versions, approvals, traffic splits, one-step rollback | Graders run on live conversations; live supervision and takeover | Home page only. The rest needs a customer login. | ISO 42001, SOC 2 Type II, HIPAA; AIUC-1 | A forward-deployed product manager and engineer, named on a public page | Custom. No public price. |
| Sierra | Simulations, which can gate a release | Voice Sims: noise, accents, interruptions, latency | Checks, merge approvals, split traffic, snapshot rollback | Monitoring and alerts; conversation review | Not found. The docs site asks for a sign-in. | SOC 2, ISO 27001, ISO 42001, PCI, HIPAA; AIUC-1 | A forward-deployed team; “agent engineers” | Outcome-based model described. No price list. |
| Decagon | Simulations, run before go-live | Accents, speed, noise, interruptions | Versions, staging and production, traffic ramp | Every conversation scored against custom criteria | Not found. The docs site redirects to a login. | SOC 2 Type II, ISO 27001, PCI, HIPAA | “Agent Product Managers” and deployment engineers | Per conversation or per resolution. No price list. |
| Kore.ai | Evaluation suites with personas and scenarios | Test calls by softphone. Noise tests not found. | Development, staging, production; rollback; approvals | Scored quality and outcome reports | Public | SOC 2, ISO 27001, PCI | Forward Deployed Engineers named; partners | Unclear |
| Google Customer Engagement Suite | Evaluations: scenario, simulation and golden test cases | Voice-mode test cases. Noise tests not found. | Fixed versions, restore, percentage traffic split | Monitoring dashboard; conversation history | Public | ISO 42001 for Google Cloud | Self-serve; general consulting | Public usage prices |
| NICE | A simulator with personas and success criteria | Not found. The docs say the simulator is text-only. | Snapshots and revert | AI observability; conversation analysis | Public | PCI, ISO 27001, SOC 2, HIPAA, FedRAMP | Professional services; certified partners | Public suite prices per agent |
| Genesys | Manual preview and a flow simulator. Automated simulation not found. | Voice preview. Noise tests not found. | Draft and publish | Performance dashboard; replay | Public | SOC 2, ISO 27001, ISO 42001, PCI, HIPAA, FedRAMP | Professional services; Forward Deployed Engineers | Public plan prices; AI by usage |
“Not found” means not found on public pages, not that the vendor lacks it. A second AI research agent gathered the rival rows from each vendor’s own pages. I checked the Cresta row against saved page text, and I have not checked the rival rows line by line. Two rivals’ trust portals and Cresta’s own could not be read by the agents, so the assurance column comes from each vendor’s other pages.
What changed in this market in 2026
Four changes that shape how an agent gets deployed now, and one thing that has not changed.
- There is now an independent standard for AI agents.AIUC-1 pairs an audit with adversarial tests on the live agent. Sierra announced it on September 17, 2026, and Cresta on November 4. Sierra, Cresta
- Testing moved from scripts to simulated customers.Five of the six rivals describe it, and three vendors including Cresta now simulate bad audio as well. Links are in the table above.
- Forward-deployed tools are being handed to customers.On October 29, 2026 Cresta opened the tools its own teams use to customer and partner developers. A playbook now has to work when the customer runs it. Self-serve model
- Some callers are now AI agents too.Cresta reports a sharp rise in personal AI agents contacting its customers, and released a model to detect them. That is a new group of test conversations. Sentry, October 12, 2026
- The decision itself is not shown.All seven vendors describe testing and release controls of some kind. On their public pages I found no go or no-go view for the customer’s executives. That is why I built one.
What is not on this page
While reading, I found a handful of places where Cresta’s public pages disagree with each other, and some smaller housekeeping.
Those are in a findings report I will send to Cresta directly. A public page is the wrong place for them.