Decagon’s Missions give AI agents weeks-long workflows

Decagon launched Missions on November 12, 2026, with AI agents built to pursue customer goals for days, weeks or months. According to Decagon, each agent can work across conversations, back-office systems and participants until it completes a defined outcome.

That longer runtime addresses a practical problem in customer operations. Decagon says Missions can “replace manual status tracking and handoff coordination with ongoing refinement of the journey.” The company expects teams to use it for revenue-bearing processes, not only question-and-answer support.

That makes the unit of automation different from a single chat response. Under Decagon’s model, an outcome may depend on a provider’s reply, payer update or customer action. The agent must preserve the thread, decide whether the next step remains necessary and resume from the right point. Decagon packages those tasks as one continuing process.

What can Missions actually do?

A team starts a Mission by stating the goal in plain language. It names the personas the agent should engage and adds any operating guidance. The agent then selects subsequent actions around that objective rather than following a fixed, linear script.

Decagon says an agent can connect separate threads involving different people. It can contact each participant through an appropriate channel, then wait when a third party must complete a step. This waiting behavior is central to the product’s pitch: customer processes often depend on outside action before the requested result can happen.

The launch examples include:

• Prior authorization for an MRI.
• Approval of a small loan.
• Lead qualification.
• Prescription refills.

Decagon describes these as multi-step journeys capable of producing meaningful revenue. For customer-experience teams, the intended market change is to hand a persistent objective to an agent instead of supervising each message and transfer separately.

Teams also decide how the agent should behave before it starts. The plain-language objective supplies the destination, personas identify the counterparties, and guidance sets boundaries for the run. After that, Decagon says the agent selects the next action from case context rather than requiring the team to script every branch in advance.

Takeaway: Missions turns a customer-service objective into work that can continue across messages, systems and people.

How does an agent move a workflow forward?

Missions extends beyond conversation through Browser Actions. According to Decagon, this feature lets an agent complete back-office work such as filling out forms. A conversation can therefore trigger and support administrative tasks within the same broader process.

The company’s prescription-refill example shows how those pieces fit. An agent first notifies the patient. It reaches out to a provider when authorization is required, and it can run claims again on a payer portal. Once the task reaches the required state, the agent confirms pickup details with the patient.

Decagon says: “A single service outcome may require customer communication, provider coordination and payer-system activity.” Missions assigns those steps to one goal-directed workflow. It does not treat customer contact, provider follow-up and payer administration as unrelated jobs.

Takeaway: Browser Actions gives the agent work to do between messages, while the Mission keeps every step tied to one outcome.

What do cases and Memory remember?

Every Mission has a case file that records progress. The record includes completed steps and new information that might change the plan. Decagon says the agent uses that case to choose its next action, schedule proactive outreach and preserve context for later work.

Before taking another step, the agent checks both the case and the customer’s Memory. The case tracks progress toward the business result. Memory holds preferences and relevant context drawn from other conversations. Together, they let the agent test whether its existing plan still fits the situation.

Decagon offers two examples of why that separation matters. In one, Memory preserves a customer’s request not to receive calls after 3 p.m. In another, it accounts for a parallel interaction in which the same customer is onboarding a new product while troubleshooting an unrelated issue. The agent can use that context without mistaking one live process for the whole relationship.

Memory also helps when the expected sequence changes. If an agent has qualified a lead and arranged a follow-up, but the customer independently schedules that meeting by email, the agent adjusts its course. It does not continue a plan that another channel has already advanced.

Takeaway: The case file remembers the job, while Memory helps the agent respect the customer’s preferences and other active matters.

How do teams debug a workflow that spans weeks?

Long automated workflows create an attribution problem. If a prescription refill is delayed, Decagon says teams may need to determine whether the fault began with the initial outreach, the provider handoff or insurance verification. Finding the interaction or handoff that needs correction becomes harder as the process stretches across more interactions.

Decagon builds every Mission as an Agent Operating Procedure, or AOP. Teams can write, edit and refine that logic. They can also inspect the agent’s decisions and test individual parts of the configuration to identify what needs improvement. Those controls aim to make a long path reviewable rather than opaque.

The company also points to Duet and Duet Autopilot as optimization tools. Decagon says they provide recommendations for the full customer journey, extending optimization beyond isolated interactions. The aim is to diagnose the sequence of actions, not merely score one exchange.

Takeaway: Missions links what an agent did with why it chose the next step, giving teams a way to test weak handoffs.

What will determine whether Missions works?

The product arrives with an enterprise focus and an open operational test. Decagon presents Missions as a way to improve customer outcomes while reducing manual coordination, but the launch description also identifies the hard part: exposing weak handoffs and improving results as interaction volume grows.

That test will depend on the durability of the underlying controls. Cases need to preserve a coherent case history. Memory needs to carry useful preferences and concurrent context without letting unrelated interactions distort the active objective. AOP testing must reveal where a multi-week path failed rather than merely describe it after the fact.

Duet and Duet Autopilot add a second layer by suggesting changes at journey level. Their value, according to Decagon, comes from recommendations that account for the full customer journey. The company’s product model is therefore not automation alone; it pairs execution with diagnosis inside the same Mission.

For buyers, the stakes sit inside Decagon’s examples. MRI prior authorization, small-loan approval, lead qualification and prescription refills all require multiple steps. A lost handoff could delay the same outcome while making the cause harder to find. Missions will need to prove that its cases, Memory and AOP controls can tell those two situations apart.

Decagon says Missions can continue for days, weeks or months. Its case, Memory and AOP controls must now show that they can expose weak handoffs and improve outcomes as interactions accumulate.