•
Install Devin Desktop from the official page, run one agent on a tiny Python lab, review the diff, and verify locally—evening-sized first session.
Devin Desktop is Cognition’s desktop home for coding agents: plan, delegate, review, and ship from one surface that includes an IDE when you need it. Official page: devin.ai/desktop.
Per Cognition’s FAQ on that page, Devin Desktop is the new name for Windsurf—an over-the-air update that keeps plans, extensions, and settings while centering an Agent Command Center (Spaces, kanban-style session views, multi-agent management) on top of the familiar IDE.
For after-work Python, treat it as: open a small Space/project, give one agent a single task, review the diff, run the script, stop.
Evening learners who want a desktop app that can run local or cloud agents and still let you read code like a normal editor. Useful if you already tried Windsurf, or if you want one place to watch agent sessions without juggling five chat tabs. Start on personal practice code; check devin.ai/desktop for download, plans, and current platform notes (Mac download is highlighted on the page; follow their instructions for your OS).
Browsing options for after-work agents? See Top 10 AI agent tools for after-work Python.
Go to devin.ai/desktop, download for your platform, and sign up / sign in. If you already use Windsurf, Cognition states Devin Desktop arrives as a standard update and migrates IDE, extensions, workflows, settings, and in-progress work—with naming/branding as the main change. Prefer their download page over third-party mirrors.
Use a dedicated folder, for example evening-python/devin-lab/, with:
# todos.py
def pending(items):
return [x for x in items if not x.get("done")]
if __name__ == "__main__":
data = [{"title": "read", "done": True}, {"title": "code", "done": False}]
print(pending(data))
Keep the workspace narrow so agents do not roam into unrelated directories.
From the agent / session UI, ask for something finishable in one sitting:
Add type hints to
pending, reject non-dict items safely, and print the count of pending titles. Do not add dependencies.
Use one session for one goal. Parallel fleets are powerful later; tonight you want a clean review loop.
Open the diff, read the function, and run:
python todos.py
Confirm the printout. If the agent added a framework or renamed the file, undo and restate “minimal change.”
Close or archive the finished session. Write one line in NOTES.md: prompt used, what you accepted, what you rejected. That note is how after-work streaks survive busy weeks.
Ask the agent to add done_titles(items) -> list[str] returning titles where done is true, then print both pending and done titles. Reject any suggestion that pulls in external packages for a list comprehension.
Agents draft code; you still own correctness. Strengthen fundamentals with SoloQuant’s Intro to Python Programming—after-work short lessons so you can steer Devin Desktop with clearer specs.