Skip to main content
Springvanta
All posts
Voice AI & IntakeMay 19, 2026 · 5 min read

Nine AI agent launches in four days: support automation hits a tipping point

HubSpot, Fin, RingCentral, SAP, and ServiceNow all shipped AI support agents during the same week. Here is what the numbers say and what it means for buyers.

By SpringVanta

Between May 12 and May 15, 2026, nine companies announced AI agents for customer support, CRM, and intake operations. Not roadmap slides. Shipped products with pricing and customer numbers.

HubSpot disclosed a 70% autonomous resolution rate on its Q1 earnings call. Fin (formerly Intercom) launched an AI agent whose only job is managing its customer-facing AI agent. RingCentral expanded its AI Receptionist to handle Shopify orders and Calendly bookings without looping in a human. SAP rolled out CX-specific agents that read ERP data and act on it. ServiceNow shipped Autonomous CRM.

Observe.AI launched a Companion Agent that coaches human agents during live calls. Zig.ai replaced traditional lead-gen tools with an AI agent team. Aline built an outbound AI agent directly into its senior living CRM. ChurnZero shipped AI-personalized email inside customer success workflows.

Nine launches. Four days. All pointing at the same market.

AI Agent Resolution Rates comparison chart

The number that matters: 20% to 70% in one year

HubSpot's Customer Agent went from resolving one in five support conversations to seven in ten over twelve months. CEO Yamini Rangan told investors the two dominant use cases are after-hours augmentation and tier-one triage. Some customers are already at 85% or higher. One burned through 5,000 included AI credits within days of turning it on and is scaling toward 300,000 credits per month.

CTO Dharmesh Shah framed the current state as a checkpoint, not a ceiling. As underlying models improve, he expects Customer Agent to move beyond tier-one support into higher-level resolution. The product just expanded into email this quarter.

This is the first time a major CRM platform has publicly reported hard resolution numbers over a sustained period with clear trajectory data. It is not a vendor demo or a pilot claim. It is a quarterly earnings figure.

Fin builds an agent to manage its agent

The company formerly known as Intercom rebranded to Fin on May 13. Two days later it announced Fin Operator, an AI system that monitors, debugs, and improves the customer-facing Fin agent.

VP of Product Brian Donohue described three roles Operator fills: data analyst (generating charts and trend reports on demand), knowledge manager (ingesting product updates and revising help articles autonomously), and agent builder (tracing misbehaved conversations to root causes and proposing fixes).

Every change Operator recommends appears as a proposal with a full diff view. Nothing goes live without a human clicking Apply. Donohue was explicit about this: zero risk mode for now. The proposal system works like a pull request in software engineering, where the AI writes the code but the human reviews it before merge.

Fin now resolves more than two million customer issues per week across 8,000 customers. It crossed $100 million in ARR, growing at 3.5x. The broader company generates $400 million. The AI agent accounts for roughly a quarter of total revenue and almost all the growth.

Operator runs on Anthropic's Claude, not Fin's proprietary Apex models. Donohue's explanation: Apex is optimized for answering customer questions. Operator's work is closer to software engineering, where Claude has been specifically tuned.

RingCentral: from answering calls to closing transactions

RingCentral's AI Receptionist (AIR) added Shopify, Calendly, and WhatsApp integrations. A customer calling after hours to book a service gets a confirmed calendar slot without reaching a person. AIR handles Shopify order status inquiries over the phone. WhatsApp gets the same AI that runs the phone channel.

Keller Interiors, an installation partner for Lowe's, deployed AIR across 33 locations. Wait times dropped from 12 minutes to 90 seconds. Customer satisfaction rose three points in four months. No new headcount.

AIR now serves more than 11,800 businesses. Joe Fahrner, VP of Growth for AI Products at RingCentral, described it as a digital team member that SMBs cannot imagine working without.

ServiceNow and the CRM that acts on its own

At Knowledge 2026, ServiceNow launched Autonomous CRM and a multimodal assistant called Otto. The core pitch: traditional CRM forces employees to act as human middleware between disconnected systems. Autonomous CRM replaces that with AI agents that execute workflows across sales, service, and fulfillment.

Rolls-Royce deployed ServiceNow's virtual agent for 12,000 employees and got a 54% deflection rate on its IT help desk. That translated to 5,000 hours of efficiency savings and 38,000 resolved incidents. Phil Priest, Head of Global Business Services at Rolls-Royce, framed the result in manufacturing terms: employees who build engines now have more time to build engines.

ServiceNow also shipped an AI Control Tower. Nanshed Barrowala, GVP of AI Products, described a common customer situation: organizations have more AI in production than they have inventoried, most of it ungoverned. He called that agentic chaos.

What this week means for businesses evaluating AI support

Three practical takeaways from the convergence:

Resolution rates are now public data. HubSpot's 20-to-70 trajectory, Fin's 2 million weekly resolutions, ServiceNow's 54% deflection at Rolls-Royce. These are not vendor claims from a controlled pilot. They are production numbers from named customers. Anyone evaluating AI support tools should ask vendors for comparable figures.

The AI agent is becoming the product. Intercom renamed itself after its agent. HubSpot's Customer Agent consumes more AI credits than every other AI feature combined. RingCentral's AIR is the growth story. The pattern: the agent is not an add-on to the platform. The platform is becoming the wrapper around the agent.

Governance is now a first-class feature. ServiceNow's Control Tower, Fin Operator's proposal system, Observe.AI's human-in-the-loop monitoring. Every major vendor is shipping guardrails alongside autonomy. The market has absorbed the lesson from early chatbot deployments: ungoverned AI creates more work than it removes.

Sources

Continue reading

← Previous

Why 84% of Developers Use AI Tools They Don't Trust

Next →

The AI Agent Accountability Gap: 80% See Risky Behavior

Read more

Like this kind of writing?

One email when something good ships — usually once or twice a month.

SubscribeAll posts