Sierra passes $400 million in ARR as RingCentral and 8x8 put AI and human agents on one supervisor view

Sierra put a number on how fast companies are buying AI agents for customer service. In a 30 Oct blog post, co-founders Bret Taylor and Clay Bavor wrote that the company passed $400 million in annual recurring revenue this week, a little over two and a half years after launch, and added more than $150 million of ARR in the third quarter alone. That is more than Sierra generated in all of last year. They also said growth is speeding up: it took six months to double from $100 million to $200 million and five months to double again to $400 million.

The post gives three reasons. Customers, many of them large and heavily regulated, are pushing their agents to do more. Agents are moving from one-off support chats to interactions that run for weeks or months and drive revenue. And Sierra's Ghostwriter tool has shortened builds, with a large US retailer with more than $40 billion in revenue going from first meeting to a live agent in 27 days this month. Sierra said it now works with almost half of the Fortune 50 and still charges for outcomes, such as a loan refinanced or a claim paid, rather than for tokens. It also said it is developing an open standard called Poppy with more than 50 enterprises, including Meta, Shopify, Stripe, and Walmart, to give businesses control over how customers' personal agents interact with their brand. The 12 Oct issue covered the Personal Agent Protocol that Sierra and Meta announced on 6 Oct with many of the same partners.

Cresta took the opposite route to the same market. On 29 Oct it opened its AI agent platform to developers who want to build customer-facing agents themselves. With Cresta Conductor, which is built on the Claude Agent SDK in collaboration with Anthropic, teams build agents through their own repository and CI/CD pipeline, then get controlled releases, A/B tests between versions, rollback, builder and reviewer permissions, and visibility into AI usage and cost. Cresta said a developer at an Australian telecom provider rebuilt a production chat agent with more than 10 subagents from another platform onto Cresta in under a week. It also launched Cresta Academy to train and certify customer and partner engineers, and quoted Atento and TELUS Digital as early partners. For buyers comparing call center AI vendors, the choice between a vendor that builds the agent for you and one that hands your engineers the tools is now an explicit one.

RingCentral and 8x8 manage AI agents like the rest of the team

RingCentral wants supervisors to manage AI agents the way they already manage people. In a 28 Oct release timed to ICMI 2026, the company said RingCX now extends live supervision, quality scoring, journey analytics, and staffing forecasts to AI agents, including its AI Receptionist (AIR) and AIR Pro products. AI conversations show up in the same live view as human ones, AI interactions are scored against the same rubric used for people, transfers carry context in both directions, and workforce forecasts plan AI and human capacity together.

The customer quote in the release explains the problem better than the feature list. Tasha McIntosh, a senior patient care manager at EVP Eyecare, said that when a request does not match what the AI agent expects, the gap "can slip through until the patient calls back frustrated." The release also adds AI Onboarding, where an administrator describes a configuration in plain language and RingCentral AI builds it through the RingCX APIs and tests the routing before go-live. RingCX is now available on mobile and inside Microsoft Teams.

8x8 made a similar argument a day later. Its 29 Oct update to the 8x8 Platform for CX is built around what it calls governed contextual memory, so a person or an AI agent can pick up where the last conversation ended. Agent Workspace now includes Customer 360, which puts a customer's interaction history in one view with sentiment and topic insights where speech analytics is in place. AI Studio agents can be built once and run across voice, SMS, WhatsApp, 8x8 Work, and web, and JourneyIQ puts every voice interaction on one timeline so teams can see where calls transfer, stall, or drop. For teams shopping CCaaS platforms, both launches are a sign that supervising AI agents is becoming a standard part of the suite, alongside workforce management and QA.

Earnings show AI is wide but shallow in contact centers

The case for those dashboards comes through in a CMSWire analysis published on 29 Oct, which went through the latest earnings reports from CX vendors. Its conclusion is that customers pick one narrow, high-volume job, such as answering missed calls, resetting passwords, or analyzing call recordings, prove the return, and then buy more. Among vendors that disclose a comparable figure, AI accounts for roughly 12 to 15 percent of revenue, depending on what it is measured against, and no vendor breaks out how much AI adds to profit.

Customer counts are climbing fast. RingCentral's AI Receptionist has more than 22,000 customers, and Zoom said its Virtual Agent customer count rose 256 percent. Share of conversations is another matter. RingCentral CEO Vlad Shmunis told analysts that "overall penetration of AI and AI-answered calls on our network is still very small percentage-wise," and Twilio CEO Khozema Shipchandler called voice AI "very early innings." CMSWire also notes the cost of getting AI live: Five9's adjusted gross margin fell to 61 percent from 63 percent after it temporarily added professional services staff to deploy AI customers faster, and Sprinklr's services business ran a negative 22 percent non-GAAP gross margin.

The analysis has a useful detail for anyone building a business case. A national fitness operator started with RingCentral's AI Conversation Expert to find out why people call, then added AIR for recurring questions and expanded its investment four times over. Conversation analytics is the other common starting point after voice self-service.

Twilio, Bandwidth, and ServiceNow report as AI traffic grows

The infrastructure underneath those agents had a strong quarter. Twilio reported third quarter revenue of $1.56 billion on 29 Oct, up 20 percent reported and 16 percent organic. Its dollar-based net expansion rate was 114 percent, up from 109 percent a year earlier, and it raised its full-year reported revenue growth outlook to 20 percent from 18 to 18.5 percent. Shipchandler said consumer AI "has accelerated AI development" and that its adoption needs fast, reliable infrastructure.

Bandwidth reported record revenue of $234 million the same day, up 22 percent from a year earlier, with adjusted EBITDA of $35 million, up 43 percent. It guided to full-year revenue of $905 million to $910 million and said its sales team added four new customer wins and expansions worth more than $1 million each. Two customer notes matter for contact centers. A fast-growing digital bank chose Bandwidth to modernize its contact center and add an AI agent, citing its native Genesys integration, and Salesforce named Bandwidth its first certified international carrier for the global rollout of its native Contact Center, covering more than 25 countries across Europe, the UK, Australia, and New Zealand. The 19 Oct issue reported that Salesforce set 16 Oct for general availability in Europe.

ServiceNow reported on 28 Oct. Subscription revenue grew 21 percent to $3,988 million, and the company raised its 2026 AI annual contract value target to $1.7 billion. Thrive, a managed services provider, now resolves 73 percent of issues at first contact using ServiceNow's CRM and AI agents, and the Bundesliga resolves 89 percent of fan inquiries at first contact after consolidating seven digital channels into one.

CX Today's service management roundup on 31 Oct read the three reports together and asked what the money buys. It noted that ServiceNow does not break out revenue from its AI service management products, that its Incident Analyzer consumes four assists per investigation, and that Salesforce's UK pricing page now lists native Contact Center at £100 per user per month, Contact Center Plus at £200, and Voice with Max Edition at £60. The piece argues that faster revenue growth does not settle who owns a failed call when a carrier, a CCaaS platform, and a CRM are all involved. Freshworks reports its third quarter on 3 Nov.

Microsoft gave CX buyers a pricing question of its own. CX Today's read of its fiscal first quarter call notes that more than 50 million agents are now registered in Agent 365, that Purview has audited more than 7.6 billion agent activities, and that Satya Nadella described the business model as expanding from per-seat subscriptions to consumption. Contact centers used to budgeting by seat will need to tie metered agent spend to resolutions, escalations, and rework.

Customers' own AI agents are already calling in

Two surveys this week suggest contact centers will soon deal with AI on both ends of the line. NiCE surveyed 118 practitioners at more than 70 of its customer organizations between 22 Sep and 10 Oct. About six in ten said they run a multi-agent ecosystem, and respondents described consumers' personal agents scheduling appointments and service agents coordinating with a logistics partner's agent. More than half said oversight, auditability, human-in-the-loop controls, or unified interaction logs are what let them expand automation. NiCE added that NiCE Cognigy can engage customers' personal AI agents over voice, digital channels, and protocols including A2A and MCP, and that it is working with Meta and other partners on the Personal Agent Protocol.

Capacity's Agentic Consumer Report, released on 27 Oct and based on a survey of 1,000 US consumers, looks at the same shift from the caller's side. Capacity found that 35 percent have used an AI agent to handle a customer service request at least once, rising to 49 percent among Gen Z and Millennial respondents. More awkward for frontline teams, 44 percent said AI had contradicted a company representative or its policies, and consumers said they were equally likely to trust either one when the two conflict. Another 23 percent said they would ask representatives to explain or cite company policies because of AI. Agents will need accurate policy answers in front of them, because more callers will arrive having already checked.

Speech-to-speech voice agents get funding and benchmarks

The vendors building the voice agents had a busy week too. Phonic released Phonic-1 on 29 Oct and announced a $27.5 million Series A led by First Round, with Lux Capital, Felicis, Definition, and Essence VC, bringing total funding to $31.5 million. Phonic's pitch is that most voice agents chain separate models together (one transcribes the caller, another decides what to say, a third speaks) and lose tone and hesitation along the way. Phonic-1 takes audio in and produces audio out, is trained on more than 50 languages, and can act while the caller is still talking so the answer is ready when they stop. The company said it powers calls on six continents for customers including Ashby, Maven AGI, and Rho, and that Phonic-1 took first place on the Artificial Analysis Speech to Speech Index.

The same day, Cartesia launched Managed Agents, a voice agent platform that bundles telephony, knowledge bases, tool integrations, and outbound campaigns. Cartesia said Managed Agents has already powered tens of millions of conversations in finance, retail, hospitality, and healthcare, and that it ranks first of 30 agents on Artificial Analysis's Speech Agent Arena, where blind human raters compare agents on tasks like booking appointments.

Regal went a different way. SiliconANGLE reported on 27 Oct that its new Goal Oriented Agents put two models on every call: a fast "actor" that handles the dialogue and a "director" that checks after each turn whether the goals for the current stage are met. In Regal's auto insurance example, a caller who describes the damage, photos, and injuries in one breath gets all three marked as collected instead of being asked for each again. Regal has raised $83 million.

Scale numbers came from Retell AI, in an OpenAI customer story published on 27 Oct. Retell serves more than 4,000 businesses, and its monthly call volume grew from 30 million in early 2026 to more than 100 million by September. OpenAI models generate responses for 80 percent of calls on the platform, and Retell reports customer resolution rates above 80 percent, including an 80.4 percent case resolution rate for Tier 1 post-sales support at Anker. Buyers comparing speech AI vendors should note that every ranking cited above is the vendor's own claim, so testing on their own calls still matters.

Also this week

McKinsey formalized its long-running collaboration with Parloa into an alliance on 28 Oct. McKinsey said most agentic customer service initiatives remain stuck in pilots, and the alliance pairs Parloa's platform with McKinsey's work on data integration, process documentation, risk and governance, operating model design, and talent strategy. Parloa chief revenue officer Chris Silver said the goal is for enterprises "to move past pilots entirely."

Genesys said on 29 Oct that Trademark Global, which sells more than 8,000 products, including Squatty Potty, through retailers such as Amazon, Walmart, Target, and Wayfair, uses Genesys Cloud Agentic Virtual Agent to check orders, troubleshoot products, and file warranty claims. Genesys reports that 42 percent of the agent's interactions finish without employee help and that it has a 91 percent success rate on warranty claims. It runs on three of Trademark Global's brands, with plans for 12 more.

Invoca launched Messaging Agents for Google Ads in beta on 28 Oct. The agents answer questions, qualify leads, and book appointments for people who message a brand from a Google search ad over RCS for Business, and they tie those conversations back to sales so marketers can optimize campaigns. Invoca said the product does not yet support conversations involving protected health information.

Zendesk shipped two AI updates for help desk admins. It announced custom summaries for tickets and calls, so admins can set the content, tone, format, and length of AI summaries and use ticket tags to apply different instructions. On 29 Oct it connected proactive messages to AI agent use cases, so a customer who replies to a proactive message lands with the right AI agent. That feature is not available to accounts on legacy AI agents with answer flows.

Telnyx now runs Pipecat voice agents on Telnyx Edge. Developers go from a repository to a callable agent with a single CLI command, and the phone number, call audio, speech to text, LLM inference, and text to speech all run on Telnyx under one API key. Two days later, on 30 Oct, Telnyx added OpenAI's open-weight gpt-oss-120b to Telnyx Inference in beta, hosted on Telnyx-owned GPUs in US regions with a 128K context window.

New in the CXSpy catalog

CXSpy added seven company pages since the last issue, bringing the catalog to 335. Phonic, covered above, joined on 31 Oct with pricing that starts at $0.07 a minute. Its customer Maven AGI, a Boston company that builds AI agents for customer support and charges per resolution, is now listed under call center AI, as is Assort Health, the San Francisco company whose voice agents answer and place patient calls for healthcare practices. Rime, which records its own conversational speech in a San Francisco studio to train voices for customer phone calls, and Sanas, best known for real-time accent translation for contact center agents, join speech AI. Modulate, which analyzes raw audio for emotion and intent rather than reading transcripts, is listed under conversation intelligence, and Pypestream, the New York company that has run AI agents for enterprise customer service since 2017, joins chatbots.

Previous issue: Gartner's 2026 CCaaS Magic Quadrant keeps the same five leaders as Microsoft brings contact center AI to rival platforms.