The cloud-native customer service company Talkdesk offers a core customer experience automation (CXA) platform withAI agent capabilitiesthat can interact directly with customers or human agents. TheCXA platform debutedin June 2025; on February 23, 2026,Talkdesk added Automation Flows(AF) capabilities.
In the announcement, Talkdesk said that AF can orchestrate AI agents and extend workflows across third-party systems for key business cases. Talkdesk also said that itsAutopilotagentic AI capabilities now operate in email, as well as in voice, chat and SMS.
The following email Q&A with Kevin McNulty, Senior Director, Product Marketing at Talkdesk, dives deeper into how Automation Flows works.
This Q&A has been lightly edited for brevity and clarity.
No Jitter (NJ): How does Automation Flows automate workflows?
Kevin McNulty (McNulty): Automation Flows is a no/low-codeorchestration layerthat automates backend processes triggered by customer events.
When a structured trigger occurs — such as a call ending, a case being created, a scheduled condition, or an API request — Automation Flows executes a defined, multi-step process.
-
Invoke integrations across systems
-
Call AI agents when intelligence is required
-
Progress through conditional paths
-
Run until a defined end state is reached
Unlike general-purpose automation tools that simply connect an event to a single action, Automation Flows governs theentire operational lifecyclethat follows a customer interaction.
It is also embedded directly in the customer experience execution layer, meaning it operates with native awareness of interaction context and works in concert with AI agents rather than sitting outside of them.
NJ: How does Automation Flows ensure that AI agents execute specificworkflows?
McNulty: Automation Flows ensures execution by actively managing the entire process from start to finish — not just triggering a series of disconnected steps.
Each workflow keeps track of where it is in the process. If a system times out or an integration fails, the workflow doesn’t collapse. It resumes from the last successful step rather than starting over or silently failing.
Built-in error handling, retry logic, and monitoring prevent issues from cascading. Workflows can also pause, wait for inputs or approvals, and then continue when conditions are met.
Before deployment, flows are validated and version-controlled to reduce instability. Once live, execution is fully observable, with step-level visibility so teams can see exactly what happened and where.
Many automation tools assume quick, one-off transactions. Automation Flows is designed for operational continuity — especially for processes that span multiple systems, multiple decisions, and longer timeframes.
NJ: What does “fully contextualized execution” mean?
McNulty: It means workflows don’t operate in isolation.
Instead of running the same predefined steps every time, Automation Flows can factor in live interaction details, customer history, account status, and information from connected systems. The workflow adjusts its path based on that context.
For example, it can treat a first-time caller differently than a long-standing customer or handle a billing issue differently if payment is already in progress in another system.
Execution reflects real-world conditions. The process responds to context, not just configuration.
NJ: Why the proviso “key business cases”?
McNulty: Automation Flows is built for the processes where the outcome truly matters — ones tied to revenue, compliance, cost exposure, or customer retention. These are multi-step, cross-system workflows where it’s not enough to trigger an action. The process has to finish correctly and the business needs to know that it did.
Examples would be things like dispute resolution, compliance validation, service recovery, or critical account updates. If those processes break halfway through or quietly fail, the impact is real.
The phrase “key business cases” is deliberate. Automation Flows focuses on the workflows where completion and visibility are business critical.
NJ: If workflow automation is just coming now, then how was CXA previously orchestrating a “network of AI agents to automate complex work” ?
McNulty: Before Automation Flows, CXA orchestratedAI agentsat the decisioning layer — coordinating how agents reasoned, invoked tools and interacted across systems. That orchestration was powerful, but it was largely confined to the agentic layer and shorter-lived execution paths.
With Automation Flows, we extend orchestration into the operational backend. This adds durable, reusable process automation that can handle more complex, multi-step, cross-system workflows — including long-running processes that need to be monitored, governed and optimized over time.
-
Before:AI agents could coordinate decisions and invoke actions across systems, but execution was limited to agent-driven flows.
-
Now:Automation Flows expands that capability into persistent, backend process orchestration — enabling more complex use cases, greater flexibility, and lifecycle management beyond the AI interaction itself.
This moves CXA beyond intelligent coordination into full operational orchestration across the enterprise.
NJ: Sounds like there’s a difference between inter-system and intra-system. Is that the case here?
McNulty: Yes. Inter-system orchestration governs coordinated execution across CRM systems, ticketing platforms, billing engines, CTI [computer telephony integration] actions and any other custom integrations. Intra-system orchestration manages logic, AI invocation, interactions and control flow within Talkdesk.
Automation Flows provides a unified execution model across both domains, purpose-built for customer experience operations rather than general IT automation.
NJ: Are any of the emerging protocols (e.g., MCP or A2A) being used?
McNulty: Talkdesk supportsModel Context Protocol (MCP)at the AI agent layer to enable structured, secure access to enterprise data and tooling. Automation Flows operates at the orchestration layer and is protocol-agnostic. It governs the multi-step backend processes triggered by interactions, APIs or AI agents.
The broader CXA architecture is designed to allow the support of enterprise interoperability and is aligned with emerging coordination models such asagent-to-agent (A2A)communication coming in the future. As multi-agent ecosystems evolve, Automation Flows provides the governed execution layer needed to ensure operational processes complete reliably and have a reporting trail across actions on several systems.
NJ: How does Autopilot use reasoning? What does reasoning mean?
McNulty: In channels like email — where requests are high-context and operationally dense — Autopilot uses reasoning to interpret unstructured input, determine required actions, verify data, update backend systems and complete resolution autonomously.
Reasoning refers to dynamic evaluation of context and system state rather than static rule trees. This enables AI agents to move beyond scripted responses into backend execution decisions.
NJ: Since AW and Autopilot are improvements to the “base” CXA, will they automatically become available to the sector-specific variants?
McNulty: Yes, they are available across all industry clouds. The orchestration engine is shared infrastructure. Industry clouds apply domain-specific workflow logic on top of the same governed execution backbone.
