AI Customer Service Agents 2026: Sierra, Decagon, Fin, Zendesk and Agentforce Compared

Customer service is where AI agents moved from demo to budget line first. The software no longer just answers questions from a help center; it looks up the order, issues the refund, rebooks the flight and closes the ticket. The market has split into three kinds of vendor, and pricing has moved from seats to resolutions. This guide explains what these agents do, compares the vendors buyers shortlist, lays out the published prices, and covers how to choose without repeating the best-known mistake in the category.

What an AI customer service agent does

A chatbot answers. An agent acts. A customer service agent reads the request in chat, email or on the phone, identifies the customer, pulls the order, account and policy from your systems, decides what to do within rules you set, does it through an API, and replies. When the case is outside its scope, or the customer asks for a person, it hands over with a summary so nobody has to start again.

• Answer: order status, delivery, account and policy questions, grounded in live data rather than a static FAQ.
• Act: refunds and credits under a limit, rebooking, address and plan changes, cancellations and retention offers, password and access resets.
• Route: classify, enrich and hand over the cases a person should handle, with the history attached.
• Assist: draft replies and surface knowledge for human agents on the cases they keep.

Gartner predicted in March 2025 that agentic AI will autonomously resolve 80% of common customer service issues by 2029. In practice the limit today is less what the models can do than which actions each company trusts an agent to take on its own. Our sister site covers that decision in AI agent use cases by business function and AI agent governance.

Three kinds of vendor

1. Agent-native platforms (Sierra, Decagon, Parloa) are built around the agent. They connect to many back-end systems, cover chat, email and voice, and sell to large consumer brands with enterprise contracts. Strongest when the agent must act across systems.
2. Helpdesk-native agents (Intercom Fin, Zendesk AI agents, Salesforce Agentforce for Service) live inside the platform where tickets already are. They learn from your existing tickets and help center on day one and publish per-resolution prices. Fastest to deploy.
3. Contact-center platforms (NiCE with Cognigy, plus the large CCaaS suites) come from voice and the contact center, with routing, workforce management and compliance built in. Strongest for voice-heavy, regulated, multi-site operations.

The vendors buyers shortlist

Sierra

Founded by Bret Taylor, former co-CEO of Salesforce and OpenAI's board chair, and Clay Bavor, formerly of Google. Sierra launched in February 2024, was valued at $10 billion in September 2025, and raised $950 million led by Tiger Global and GV in May 2026 at a $15.8 billion valuation, the largest in the category. It reports serving 40% of the Fortune 50; public customers include ADT, Sonos, WeightWatchers and Casper. Sierra is the best-known advocate of outcome-based pricing, where the customer pays when the agent resolves an interaction, under enterprise contracts rather than a public rate card. Best fit: large brands that want a vendor to build and run a branded agent across chat and voice.

Decagon

Founded by Jesse Zhang and Ashwin Sreenivas. Decagon raised $250 million in January 2026, co-led by Coatue and Index Ventures, at a $4.5 billion valuation, up from $1.5 billion in June 2025. Customers named at the raise include Avis Budget Group, Hertz, Block, Affirm, Chime, Deutsche Telekom and Grubhub. One agent engine covers chat, email and voice, and the pitch is that support and operations teams, not only engineers, configure the agent's workflows and guardrails. Pricing is by contract. Best fit: fintech, travel and consumer subscription companies with many account actions to automate.

Intercom Fin

Intercom's agent is the clearest published price in the market: from $0.99 per outcome, on any Intercom plan or as a standalone agent. Intercom seats start at $29 per seat per month for Essential (discounted for new customers), $85 for Advanced and $132 for Expert, billed annually. Fin can run on top of other helpdesks, including Salesforce, so it is not limited to Intercom shops. Best fit: SaaS and digital businesses that want a fast start and a predictable unit price.

Zendesk AI agents

Zendesk includes a small allowance of automated resolutions in its plans: 5 per agent per month on Support Team and Suite Team, 10 on Suite Professional. Beyond that it charges $1.50 per automated resolution on a commitment, or $2.00 pay-as-you-go. Suite plans list at $55 (Team) and $115 (Professional) per agent per month billed yearly. Best fit: companies already on Zendesk who want resolution pricing without a second vendor.

Salesforce Agentforce for Service

Agentforce runs inside Service Cloud with access to the CRM record. Customer-facing agents are sold at $2 per conversation on pre-purchase, or through Flex Credits at $500 per 100,000 credits, where a standard action draws 20 credits. Employee-facing Agentforce add-ons list at $125 per user per month. The pricing has changed several times since the 2024 launch, so model your own volumes with a Salesforce account executive. Best fit: Service Cloud customers whose case data and workflows already live in Salesforce. Our sister site reviews the Agentforce Specialist certification.

Ada

Ada, one of the older AI customer service vendors, covers messaging, email and voice in dozens of languages. It does not publish a rate card; most customers buy on conversation-based pricing, with resolution-based pricing for some enterprises. Best fit: multilingual consumer brands that want a managed platform with a long track record.

Parloa

Berlin-founded Parloa raised a $350 million Series D led by General Catalyst in January 2026 at a $3 billion valuation, and has set up a US headquarters in New York. Customers include Allianz, Booking.com, HealthEquity, SAP, Swiss Life and TeamViewer. It sells an AI Agent Management Platform and is strongest in phone-heavy service in insurance, travel and financial services. Pricing is by contract.

NiCE Cognigy

NiCE, one of the largest contact-center software vendors, closed its $955 million acquisition of Düsseldorf-based Cognigy in September 2025 and is combining Cognigy's conversational and agentic AI with its CXone Mpower platform. Best fit: large contact centers already on NiCE, or buying routing, workforce management, quality management and AI agents from one vendor.

Also on many shortlists: Kore.ai, Google's Customer Engagement Suite, Amazon Connect and Microsoft Dynamics 365 Contact Center for companies standardized on those clouds, and Cresta for real-time assistance to human agents.

AI customer service agents compared

The per-resolution math

Outcome pricing is easy to explain and easy to misjudge. Three checks before you sign:

• Run your volume through it. An agent resolving 20,000 contacts a month at $1 to $2 per resolution costs $20,000 to $40,000 a month in usage alone, before seats, integration and the team that maintains the agent's knowledge and policies. Compare it with your fully loaded cost per human-handled contact, not with the license line.
• Read the definition of a resolution. Vendors define it differently. A definition that counts any conversation where the customer did not ask for a human rewards an agent that wears customers down. Ask for resolution to be confirmed by the customer or by the absence of a repeat contact within a set window.
• Price the actions, not just the answers. The value is in what the agent can do: refunds, changes and cancellations. Each needs an integration and a policy on limits. Budget that work; it is usually larger than the license in year one.

For the wider cost picture of AI spend, see our AI FinOps guide.

The Klarna lesson

In 2024 Klarna said its AI assistant was handling most customer chats and doing the work of 700 agents. In May 2025 CEO Sebastian Siemiatkowski told Bloomberg that the cost focus had gone too far and that "what you end up having is lower quality." Klarna began hiring human agents again and promised that "there will always be a human if you want." The lesson is not that agents fail. It is that containment rate is the wrong target on its own. Measure satisfaction and repeat contacts on agent-resolved cases separately, and keep an easy path to a person.

How to choose

1. Start from your platform. If you run Intercom, Zendesk or Service Cloud, pilot the native agent first. It already has your tickets and knowledge base, and you learn what your contacts need before paying for a bigger build.
2. List the top 20 contact reasons and the action each needs. If most are answers, a helpdesk-native agent is enough. If most need actions across billing, order and booking systems, shortlist Sierra and Decagon.
3. Test on your own transcripts. Give every vendor the same few hundred real conversations, including angry, ambiguous and out-of-policy ones, and grade the outcomes yourself. Our sister site has a guide to testing and evaluating AI agents.
4. Decide voice now. If phone is a large share of contacts, weigh Parloa, NiCE Cognigy, Sierra and Decagon, whose voice agents are core products rather than add-ons.
5. Set autonomy limits before launch. Refund caps, which actions need human approval, and when the agent must hand over. Security deserves its own review because customer-facing agents read untrusted input; see AI agent security vendors.
6. Measure the right things. Cost per resolved contact, first-contact resolution, satisfaction on agent-handled contacts and repeat contacts within seven days, reported next to containment.

Sources

• Built In, Agentic customer experience platform Sierra raises $950M (May 5, 2026); Value Add VC, Sierra AI valuation 2026.
• SiliconANGLE, Decagon AI raises $250M at $4.5B valuation (Jan 28, 2026).
• Intercom, Pricing; Zendesk, Pricing (both accessed Oct 2026).
• Salesforce Break, Agentforce pricing tiers explained (Jul 17, 2026).
• PR Newswire, Parloa valued at $3 billion with $350M Series D (Jan 15, 2026).
• Dallas Innovates, NiCE closes $955M acquisition of Cognigy (Sep 2025).
• Chatbase, Ada CX pricing (third-party summary of Ada's terms).
• Maginative, Klarna dials back its AI customer service strategy (May 2025).
• Gartner prediction (March 2025) on agentic AI resolving 80% of common service issues by 2029, discussed by Maven AGI.

Related reading: AI agent use cases by business function on aistrategy.guide, our enterprise AI agent platforms guide, and agentic AI courses on CTAIO.

What is an AI customer service agent?

An AI customer service agent is software that handles customer requests end to end: it understands the request, looks up the customer, order and policy in your systems, takes an action such as a refund, rebooking or address change, replies, and hands over to a human with a summary when it cannot or should not resolve the case. That ability to act is the difference from an FAQ chatbot, which only answers questions from a knowledge base.

What is the best AI customer service agent?

It depends on your stack and how much the agent must do. If you already run Intercom, Zendesk or Salesforce Service Cloud, start with Fin, Zendesk AI agents or Agentforce, because they read your existing tickets and knowledge base on day one. If the agent must act across many back-end systems and channels, including voice, at enterprise scale, Sierra and Decagon are the agent-native leaders. For voice-heavy contact centers, look at Parloa and NiCE Cognigy.

How much do AI customer service agents cost?

Published prices in October 2026: Intercom Fin from $0.99 per outcome; Zendesk $1.50 per automated resolution on a commitment or $2.00 pay-as-you-go beyond the 5 to 10 included per agent per month; Salesforce Agentforce $2 per conversation for customer-facing agents, or Flex Credits at $500 per 100,000 credits. Sierra, Decagon, Ada, Parloa and NiCE Cognigy price by contract. Add the helpdesk seats, integration work and the people who maintain the agent's knowledge and policies.

Is per-resolution pricing cheaper than per-seat?

It aligns cost with value, but it is not automatically cheaper. At $1 to $2 per resolution, an agent resolving 20,000 contacts a month costs $20,000 to $40,000 a month in usage. Compare that with your fully loaded cost per human-handled contact. Also check how each vendor defines a resolution: some count any conversation where the customer did not ask for a human, which can reward an agent for frustrating customers into leaving.

Can AI agents replace customer service teams?

Not fully, and the best-known attempt is a warning. Klarna said in 2024 that its AI assistant was doing the work of 700 agents. In May 2025 its CEO Sebastian Siemiatkowski told Bloomberg the cost focus had gone too far and produced lower quality, and Klarna began hiring human agents again while promising customers there will always be a human if they want one. Agents absorb volume; people still handle the complex, emotional and high-value cases.

What is the difference between Sierra and Decagon?

Both are agent-native platforms built for large consumer brands, both cover chat, email and voice, and both connect to back-end systems to take actions. Sierra, founded by Bret Taylor and Clay Bavor, is the larger company ($15.8 billion valuation after a $950 million round in May 2026) and is known for outcome-based pricing. Decagon, founded by Jesse Zhang and Ashwin Sreenivas, raised $250 million at a $4.5 billion valuation in January 2026 and emphasizes letting non-engineers configure agent logic. Run both on your own transcripts before choosing.