How do I define a schedule for a Managed Deep Agent?

Create a file in the 'schedules' directory and define the cron expression, time zone, and the specific prompt you want the agent to execute.

Does scheduling an agent prevent it from answering manual questions?

No, the agent continues to function as a standard agent; it can handle both scheduled recurring tasks and ad-hoc user queries.

Where can I see my scheduled agents after deployment?

After running the deploy command, you can view and inspect your crons and prompts in the 'crons' tab of your deployment within the LangSmith UI.

Scheduling Managed Deep Agents with LangChain and LangSmith

Learn how to use cron syntax to automate recurring AI tasks and deliver proactive reports to Slack using LangChain's Managed Deep Agents and LangSmith.

Imagine an AI assistant that doesn't just wait for you to ask a question, but wakes up every Monday morning to deliver a curated business report directly to your Slack channel. This shift from reactive chatbots to proactive agents transforms how teams interact with their data.

⚡ TL;DR: LangChain's Managed Deep Agents now support schedules via cron syntax, enabling agents to automatically execute tools and deliver recurring reports to channels like Slack. By deploying these configurations to LangSmith, developers can create proactive agents that handle both scheduled summaries and ad-hoc queries simultaneously.

The Simple Version

Most AI tools are like a vending machine: you put in a prompt, and you get a result. You have to initiate every single interaction. If you want a weekly report, you have to remember to ask for it every Monday.

Schedules for Managed Deep Agents turn that vending machine into a personal assistant. Instead of waiting for your command, the agent follows a calendar. It knows exactly when to wake up, which tools to use to gather data, and where to send the final summary.

🎯 Think of it this way: It is the difference between a Search Engine (where you look for info) and a Newsletter (where the info finds you). You are essentially creating a personalized, AI-powered newsletter that queries your live business data in real-time.

How It Actually Works

The process of scheduling an agent is handled through a combination of local configuration files and the LangSmith deployment platform. It moves the logic of "when to run" out of the application code and into the agent's configuration.

Defining the Schedule

To set up a schedule, developers create a new file within a specific schedules directory. Inside this file, they define the cron expression, the target time zone, and the prompt that the agent should execute when the timer hits.

For example, in the demonstration, Nathan Drezner sets up a prompt to "write the Monday pipeline for the sales team" with a specific structure focusing on risk rather than volume. This allows the agent to act as an analyst, filtering Salesforce data to find deals that may be at risk.

Connecting the Delivery Channel

A schedule is only useful if the output reaches the right people. LangChain utilizes a "channel system" to bridge the gap between the agent's internal logic and external communication tools. By configuring a Slack channel for the agent, the output of the scheduled run is automatically pushed to a specific conversation.

🎯 Think of it this way: The cron job is the alarm clock, the prompt is the to-do list, and the Slack channel is the delivery mailbox.

Deployment to LangSmith

Once the files are configured locally, the developer uses a deploy command. This publishes the agent to LangSmith and configures the cron job on the server side. As Drezner explains, "this will publish the agent to LangSmith, and it'll configure the cron."

Once deployed, the schedules are visible in the LangSmith UI under a dedicated "crons" tab. This gives developers a centralized place to inspect the timing and the prompts associated with every automated task.

Technical Deep Dive

From an architectural perspective, Managed Deep Agents operate on an agent server that decouples the execution trigger from the user interface. When a scheduled event occurs, the server initiates a new thread. This is critical because it ensures that the scheduled report doesn't clutter the history of a user's active chat session, while still allowing the agent to leverage its full set of tools and metadata.

The Role of MCP Servers

In the provided example, the agent interacts with Salesforce. While the video uses a mock connection, Drezner notes that in production, you would connect to Salesforce using an MCP (Model Context Protocol) server.

For context, MCP is an open standard (introduced by Anthropic) that allows developers to build a single "connector" for a data source that can be used across different LLMs and agent frameworks. Instead of writing a custom API wrapper for every single agent, an MCP server provides a standardized way for the agent to discover and call tools, making the integration with complex CRM systems like Salesforce more robust and portable.

Hybrid Execution Model

A key technical detail is that scheduling does not lock the agent into a single mode of operation. The agent maintains a hybrid execution model. It can handle: 1. Asynchronous Triggers: The cron-based events that fire independently. 2. Synchronous Triggers: Ad-hoc user queries via tagging the agent in a channel.

This is achieved by maintaining a shared set of instructions and tool definitions. Whether the trigger is a clock or a human, the agent uses the same underlying logic to fetch data and generate responses. As Drezner puts it, "just because an agent is configured on a schedule doesn't mean that's the only thing that it can do."

Why This Matters

The transition to scheduled agents solves a major pain point in enterprise AI: prompt fatigue. In most companies, the value of an LLM is locked behind the user's willingness to write a good prompt. By automating the prompt via a schedule, the AI becomes a proactive participant in the business workflow rather than a passive tool.

Compared to traditional Robotic Process Automation (RPA), which follows rigid, brittle paths, scheduled Deep Agents are flexible. If the structure of a Salesforce report changes slightly, a well-prompted agent can often adapt and still find the "at-risk deals" without a developer needing to rewrite the integration code.

Ultimately, this enables a "management by exception" workflow. Instead of managers spending hours digging through dashboards to find problems, the agent highlights the exceptions—the risks and the anomalies—and delivers them directly to the team's communication hub every Monday morning.