Pega Customer Service is moving toward a governed workflow layer beneath the contact center, with Infinity 26 adding MCP connectivity, agentic service capabilities, and Predictable AI pricing while leaving core CCaaS functions to partners.
Pega’s 2026 roadmap is built around something customers have been saying for a while: impressive AI doesn’t count for much when the job’s still unfinished. A February 2026 Pega and YouGov survey of 4,748 adults in the US and UK found that 46% rarely or never get a successful result from AI-powered service. Some 77% said humans do better, and just 2% wanted GenAI handling everything.
That explains the direction of Pega Customer Service. Infinity 26 reached general availability on July 14, 2026, with Pega positioning governed workflows, external-agent connectivity through Model Context Protocol (MCP), and Predictable AI pricing as the route from agent experiments to production work.
What’s interesting is what Pega isn’t trying to own. Its big contact center play in 2026 puts Amazon Connect or another CCaaS platform around the conversation, while Pega controls the persistent case, business rules, approvals, and back-office action underneath it. That could remove a lot of brittle handoffs in a complex enterprise. It could also leave buyers governing another expensive layer in an already crowded stack.
TL;DR: What’s Pega’s 2026 Customer Service Bet?
- Pega Infinity 26 is now generally available, and its customer service roadmap centers on governed case execution rather than a full CCaaS replacement.
- Predictable AI shifts away from metered Pega-managed token charges, but buyers still need the contract definition of a billable or resolved case.
- MCP lets authorized third-party agents discover and execute Pega processes, while Pega can invoke outside agents inside a workflow step.
- Pega’s Amazon Connect positioning is unusually clear: AWS owns the contact center layer, while Pega provides the workflow engine behind AI and human service interactions.
What Changed in Pega Customer Service With Infinity 26?
Infinity 26 changes the commercial and connectivity logic around Pega Customer Service. The July 2026 release brings its current agentic architecture into general availability, while Pega’s June roadmap makes the priority clear: governed service processes that can use AI without handing an open-ended model control of the customer journey.
| 2026 change | What it adds | Buyer read |
|---|---|---|
| Predictable AI | Pricing tied to completed work rather than metered Pega-managed tokens | Easier unit to budget, but billing definitions still need contract detail |
| Runtime MCP | Authorized external agents can discover and execute Pega processes | More model choice without giving up Pega’s case controls |
| Agentic Assignment | AI can chase missing information or approvals | Useful for work that stalls between conversations |
| Document Agent | Splits, classifies, scores, and triages documents | Practical for claims, disputes, onboarding, and regulated service |
| Infinity Studio | AI-first development environment with coding-agent support | Could speed delivery, though live-service impact depends on implementation |
The release adds runtime MCP, agentic assignments, a Document Agent, and Infinity Studio. Pega Academy material for Customer Service ’26 also shows an Agentic Messaging interface where a self-service agent can access customer data and submit requests. Agentic Email was preview-only in 25.1.2 with GA planned for Infinity 26, so buyers should verify entitlement in their own environment.
There’s continuity too. Pega Agent Experience already brought third-party agents into multi-step work. MCP standardizes that direction rather than inventing a new one. Clearly, Pega is investing in the case behind the contact, not native voice, WFM, or quality management.
Key Takeaways
- Infinity 26 reached GA on July 14, 2026, but feature-level availability and licensing still need checking.
- MCP expands an orchestration strategy Pega was already pursuing rather than creating one from scratch.
- The product investment is concentrated around governed case execution and agentic workflows, not a wholesale CCaaS rebuild.
What Does Pega’s Predictable AI Pricing Mean for Contact Center Buyers?
Pega Predictable AI replaces metered Pega-managed token charges with pricing tied to completed work, which is a much easier unit for a service buyer to understand. Pega’s June launch material described a flat charge per completed “case,” while its July 14 Infinity 26 GA announcement talked about predictable costs aligned with completed work and resolved cases. Procurement teams need that commercial definition written down.
Pega says heavy reasoning happens at design time, where Blueprint and Infinity Studio define the workflow. At runtime, lighter AI routes the customer into an approved process and calls a larger model only for bounded tasks. Its calculator argues that LLM-led orchestration can cost five to 20 times more per run because the model repeatedly rereads context and chooses the next step.
The public calculator still isn’t a price list. Its worked example uses an illustrative $0.88 per case at roughly one million annual cases over three years. Buyers need answers for reopened work, abandonment, human completion, multiple workflows, and outside-model failures. A closed workflow and a customer who doesn’t call back tomorrow aren’t automatically the same outcome.
| Scenario | Question to put in the contract |
|---|---|
| Customer abandons the journey | Is unfinished work billed? |
| Employee completes the final stage | Does it still count as billable AI work? |
| Case is reopened | Is there another charge? |
| One request spawns several workflows | How many billable cases appear? |
| External model fails | Who absorbs the model and recovery cost? |
Key Takeaways
- Predictable AI removes metered Pega-managed token charges, but the billable case definition belongs in the contract.
- Pega’s calculator is illustrative, so don’t build a business case around the default $0.88 example.
- Measure durable resolution and repeat contact alongside workflow completion, or a neat invoice can hide a messy customer outcome.
Learn more about the growing costs of AI-powered service here.
What Does Pega’s MCP Support Let External AI Agents Do?
Pega’s MCP support lets an authorized third-party agent discover and execute approved Pega processes, and it lets a Pega workflow call an outside agent for a defined task. Pega’s June 8 announcement names Anthropic Claude, Google Gemini, OpenAI, and AWS AgentCore as examples. That gives enterprises model choice without giving every model unrestricted access to the systems behind a customer case.
The most valuable part is the two-way pattern. A bank could use one outside agent to interpret a dispute document, then let Pega apply account rules, request an approval, and carry the case forward. Swapping the specialist model shouldn’t require rebuilding the entire service process.
Pega also added an Agentic Assignment capability for work that stalls because someone needs information or approval, plus a Document Agent for splitting, classifying, scoring, and triaging files. Those are practical additions for claims, disputes, onboarding, and regulated service, where the awkward work often happens between conversations rather than during them.
MCP solves connection problems. It doesn’t write the governance policy for you. Pega AI Lab Director and Lead Scientist Peter van der Putten told our team on June 16 that simply throwing a model at a problem and expecting it to sort itself out is “magical thinking.” Buyers still need agent identities, permissions by step, model-version history, audit records, and a decent plan for when an external model fails or starts behaving strangely.
Key Takeaways
- MCP opens approved Pega workflows to outside agents and lets Pega invoke external agents inside a case.
- The model can be replaceable while the governed service process stays inside Pega.
- MCP solves connectivity, not accountability. Identity, permissions, auditability, and recovery still need explicit controls.
Is Pega Replacing the CCaaS Layer or Orchestrating Beneath It?
Pega’s 2026 answer is much closer to “orchestrating beneath it.” At PegaWorld on June 8, Pega and AWS described Pega Customer Service as the workflow engine powering Amazon AI Agents inside Amazon Connect. AWS keeps the cloud contact center layer, while Pega gives AI agents and customer service representatives the same processes, rules, and expected outcomes behind the interaction.
A short customer call can create a long tail of work, which is where the split gets interesting. Amazon Connect can own voice, routing, and digital conversations, then Pega can carry the case through approvals, policy logic, and follow-up after the call ends. Pega isn’t tied to AWS either, with integrations spanning Five9, Genesys, Cisco, Avaya, and Talkdesk.
Its older customer evidence gives the workflow story some weight: Primerica reports lower call volume and shorter testing cycles, NHS 24 cut handling time by 10 minutes, and MetLife Mexico accelerated policy issuance. Just don’t mistake that for Infinity 26 proof. The newer agentic layer still needs its own evidence.
| Evidence | Reported result | What it supports | Caveat |
|---|---|---|---|
| Primerica | 30% lower call volume via self-service; 70% shorter testing cycles | Workflow + unified desktop case | Pega-published; predates Infinity 26 |
| NHS 24 | 10 minutes off average patient handling time | Process support around high-volume service | Pega/customer evidence; not an Infinity 26 agentic result |
| MetLife Mexico | 40% lower policy issuance time; fewer missing-info stalls | Back-office workflow effect on service | Supports orchestration, not new MCP pricing claims |
Key Takeaways
- Pega’s own PegaWorld framing puts Customer Service underneath Amazon Connect as the workflow engine, not above it as a CCaaS replacement.
- The strongest customer evidence shows value when service depends on process and back-office work, not just faster conversations.
- Existing case-study outcomes predate much of Infinity 26, so they support the architecture more than the newest agentic claims.
What Is Missing From Pega’s Customer Service Roadmap?
The obvious omissions are native voice infrastructure, workforce forecasting, quality management, recording, supervisor intervention, and production observability for mixed AI-agent estates. Pega’s 2026 announcements spend far more time on workflow execution, agent connectivity, development, and governance. That looks deliberate rather than accidental, and it tells buyers where Pega wants to compete.
Pega can avoid rebuilding telephony, media handling, WFM, and recording while plugging into established CCaaS platforms. The customer still experiences one service journey, though, so somebody has to own latency, broken handoffs, context loss, and the ugly edge cases that cross vendor boundaries.
Customer Engagement Studio reinforces the broader philosophy without changing that contact center gap. It sits in Customer Decision Hub, not Customer Service, and coordinates Pega and third-party agents across marketing operations. Pega’s current CDH ’26 material now lists the Studio as part of the release. That’s evidence of a wider orchestration-first strategy, but it doesn’t add native routing, workforce tools, or a new voice stack to Pega Customer Service.
That control-first instinct shows up outside customer service too. In April, Pega launched a FedRAMP High-authorized Blueprint environment for US government agencies. That tells buyers something about where Pega is putting its effort. It doesn’t mean every Customer Service deployment automatically gets the same controls.
Key Takeaways
- Pega is leaving several core contact center operating functions to CCaaS and workforce partners.
- Customer Engagement Studio supports the orchestration thesis, but it belongs to Customer Decision Hub rather than Customer Service.
- Governance is clearly a Pega-wide priority; buyers still need product-specific evidence for their own service deployment.
How Does Pega Compare With Salesforce and ServiceNow, and What Should Buyers Test?
Pega, Salesforce, and ServiceNow can all put an AI agent in front of the customer. The interesting bit starts when that agent has to change something that actually matters. Pega hands the request to a persistent case. Salesforce leans into CRM data and actions. ServiceNow is building a stronger control layer around agents working across the enterprise.
| Buyer question | Pega | Salesforce | ServiceNow |
|---|---|---|---|
| What carries the work? | Persistent case and workflow | CRM data, knowledge, service actions | Now Platform workflows |
| Where does it look strongest? | Claims, approvals, regulated processes | Service centered on Salesforce data | Mixed-agent governance across systems |
| What needs challenging? | Pricing, migration, partner joins | Depth beyond Salesforce context | Cost and operational weight |
That comparison helps explain Pega’s Forrester result. In the Q1 2026 Customer Service Solutions Wave, Pega says it received the highest available score in 16 of 31 criteria, including case management, CSR assistance, process automation, and AI agents. Those are exactly the areas its roadmap keeps emphasizing.
Depth is useful until it becomes another pile of complexity. So don’t just watch the happy-path demo. Ask Pega to show a reopened case, a dead external model, a customer demanding a person, and a failed handoff from Amazon Connect.
Key Takeaways
- Pega’s edge is the governed case that keeps working after the conversation ends.
- Forrester supports that workflow depth, but buyers still need to test the joins around it.
- Measure whether Pega finishes customer work cleanly, not how impressive the agent looks in isolation.
Who Is Pega Customer Service Best Suited to in 2026?
Pega Customer Service looks strongest for large enterprises where a customer interaction kicks off consequential work across people, systems, approvals, and policy. Claims, disputes, account changes, healthcare requests, and regulated decisions are the obvious fit because the service problem doesn’t end when the chat window closes or the customer hangs up.
Infinity 26 helps with that position. MCP makes it easier to bring outside agents into governed processes. Predictable AI gives procurement a more recognizable commercial unit than raw token consumption. The Amazon Connect story also gives buyers a clearer architecture: keep the channel and telephony platform you already trust, then use Pega to carry the case to a finished outcome.
The weakness sits in the same design. If the enterprise still needs separate platforms for voice, WFM, quality, recording, routing, and AI estate monitoring, Pega can become one more strategic layer to license and govern. The value only appears if its workflow engine removes more friction than it adds.
Pega isn’t racing to own every contact center component. It’s betting that the valuable part of agentic customer service is controlling what happens next. For buyers with complex, long-running service work, that’s a credible bet. For buyers primarily chasing an all-in-one CCaaS consolidation, it’s a much harder sell.
FAQs
Is Pega Infinity 26 available now?
Yes, as of July 14, 2026. Just don't read “GA” as “everything's turned on for everyone.” Customer Service capabilities, licensing, model access, and regional availability can still vary. Get the exact matrix for your Pega Cloud configuration before you score anything as available.
What does Pega count as a completed or resolved case?
Pega's June Predictable AI material described pricing around completed cases, while the July Infinity 26 announcement also used resolved-case and completed-work language. Public material still doesn't answer every billing edge case. Buyers should define reopened work, abandonment, human completion, failed external models, and multi-workflow requests in contract language before forecasting costs.
Does Pega charge for AI tokens?
Pega says Predictable AI removes metered charges for Pega-managed model tokens and ties costs to completed work instead. That doesn't make the full stack token-free or cost-flat. External models, CCaaS services, platform licenses, partner delivery, data services, and workflow maintenance can still create separate costs that belong in the total business case.
Can external AI agents execute Pega workflows?
Yes, when they're authorized. Pega's MCP support allows compatible third-party agents, including agents built with Claude, Gemini, OpenAI, and AWS AgentCore, to discover and run approved Pega processes. Pega workflows can also call an outside agent for a defined step. The connection is open; the process and permissions still remain governed inside Pega.
Does Pega Customer Service replace a CCaaS platform?
Usually, that isn't the 2026 architecture Pega is emphasizing. Pega's Amazon Connect material positions Customer Service as the workflow engine beneath the contact center layer. A CCaaS vendor can continue to handle voice, routing, and digital contacts while Pega manages the case, rules, approvals, enterprise data, and work that continues after the interaction.
