Agentic Coding with Windsurf: Cascade, Planning, and Devin Cloud Agents

Class Duration

14 hours of live training delivered over 2-3 days to accommodate your scheduling needs.

Student Prerequisites

Professional software development experience in at least one language
Working knowledge of Git and GitHub (issues, pull requests, branches, reviews)
Windsurf (now Devin Desktop) installed, with access to Devin cloud agents
Comfort working in an AI-first editor
Familiarity with Windsurf's core features is helpful

Target Audience

Software engineers, tech leads, and platform/DevEx teams who want to move beyond AI autocomplete and delegate engineering work to Windsurf agents. Ideal for teams adopting agentic workflows across the editor and the cloud—running Cascade locally, launching Devin cloud agents from GitHub, Slack, and Linear, and organizing parallel work in the Agent Command Center. This is a practitioner course focused on agentic development with Windsurf rather than a survey of every feature.

Note: Windsurf is now Devin Desktop (Cognition). This course uses the current Cascade + Devin cloud agent model; "Windsurf" and "Devin Desktop" refer to the same editor.

Description

Windsurf pairs Cascade—the in-editor agent—with Devin, the cloud agent that runs work asynchronously, all coordinated through the Agent Command Center. This course teaches developers how to put both to work on real tasks—planning multi-step changes, editing across many files, running commands, and iterating until the job is done. You will learn to launch work outside the editor by handing tickets to Devin from GitHub, Slack, Linear, or the web, then track and organize that work in the Agent Command Center with Spaces and a Kanban view as agents work on branches and open pull requests. We cover Cascade's planning mode and tool calling, grounding agents with Memories and Rules and MCP servers, running multiple agents in parallel, and closing the loop with agentic PR review. Throughout, we emphasize supervision, security, and cost so teams can adopt agentic coding responsibly.

Learning Outcomes

Explain Windsurf's agentic surfaces—Cascade in the editor and Devin cloud agents—and choose the right one per task.
Drive Cascade for multi-step, multi-file tasks: Code mode, tool calling, and dependency installation.
Use Cascade's planning mode to draft and edit a multi-step plan before execution.
Launch Devin cloud agents from GitHub, Slack, Linear, and the web.
Organize and manage agents in the Agent Command Center with Spaces and the Kanban view.
Understand how Devin runs async, works on a branch, and opens pull requests.
Write agent-ready issues with clear scope and acceptance criteria, and iterate through PR review.
Ground agents in project context with Memories, Rules, Workflows, and MCP servers.
Apply agentic PR review and route findings back to an agent for fixes.
Establish security, governance, and cost controls for agentic development at team scale.

Training Materials

Comprehensive courseware is distributed online at the start of class. All students receive a downloadable MP4 recording of the training.

Software Requirements

Windsurf (Devin Desktop) installed, access to Devin cloud agents, a GitHub account with a repository and the GitHub integration connected, optional Slack/Linear integrations, and Git.

Training Topics

Foundations of Agentic Coding

What "agentic" means: plan, act, observe, iterate
Windsurf's agentic surfaces: Cascade (in-editor) and Devin (cloud), coordinated by the Agent Command Center
The supervision spectrum: Chat mode → Code mode → autonomous cloud work
Where agents excel and where to keep a human in the loop

Cascade in the Editor

Code mode (create and modify) vs. Chat mode (questions and exploration)
Tool calling: Search, Analyze, Web Search, MCP, and terminal commands
Autonomous package detection and dependency installation
Named checkpoints, reversible reverts, and linter integration
Voice input and real-time awareness of your actions

Workflows and Multi-Step Automation

Defining Workflows for repeatable, multi-step tasks
Todo lists for complex work
Chaining tool calls within a single task
One-click app deployment from Cascade

Planning and Decomposition

Planning mode: a specialized planning agent refines the long-term plan
Drafting and editing a multi-step plan before code changes
Turning a goal into agent-sized tasks
Acceptance criteria the agent can verify against

Grounding Agents in Context

Memories that persist context across sessions
Rules to customize agent behavior
MCP servers for databases, GitHub, Linear, and browser automation
Context management strategies for large codebases

Launching Devin Cloud Agents

Handing tickets to Devin from GitHub, Slack, Linear, and the web
Starting async work without leaving your workflow
Choosing between Cascade (local) and Devin (cloud) for a task
Tracking progress through session logs

The Agent Command Center

Managing all your agents from one place
Spaces for organizing agent work
Kanban view for tracking tasks across stages
Multi-agent management and orchestration

From Issue to Pull Request

Writing agent-ready issues and tickets (GitHub, Linear)
Reviewing Devin's branch and pull request
Iterating with the agent through comments and re-prompts
Handing off between Cascade and Devin mid-task

Agentic Code Review

Devin reviewing pull requests for correctness and project fit
Routing review findings into follow-up fixes
Combining human review with agent-generated follow-ups
Quality gates before merge

Remote and Multi-Agent Workflows

Running multiple Devin agents in parallel from the command center
Coordinating Cascade and cloud agents on one effort
Parallelizing independent tasks and merging results
Avoiding conflicts across parallel agents

Security, Governance, and Cost

Repo access and environment scoping
Memories and Rules governance for consistent behavior
Privacy and enterprise controls
Usage and cost management for cloud agent runs

Workshop

Launch a Devin cloud agent from GitHub or Slack and ship its PR to merge
Cascade lab: planning mode → multi-step build
Agentic PR review → fix agent
Agent Command Center multi-agent exercise
Q&A session