AI Coding Workflows: From Cloud to Local
by DeepLearning.AI
Rebuild the same Python app five times — cloud agent to fully local — and measure what each setup costs.
Overview
Fifteen short video lessons totalling 1 hour 32 minutes rebuild the same Python application under progressively more local coding-agent setups, so every change is measured against one fixed baseline. The sequence is deliberate: Claude Code alone, then Claude Code with subagents that split the job into focused tasks, then those same subagents driven by a cheaper model such as Haiku, then a switch to OpenCode — an open-source coding agent — pointed at OpenRouter so the inference provider becomes a swappable variable, then a hybrid setup with LM Studio serving a local Gemma 4 12B, and finally a fully local run on Qwen 3.5 27B. At each step the course records cost, speed, token usage and, crucially, what data leaves the machine; two optional readings supply a glossary and the evaluation metrics used. Instructor Paul Everitt is a JetBrains developer advocate, and JetBrains' own write-up frames the course around sovereign AI, privacy and cost rather than benchmark chasing, describing it explicitly as a gentle introduction to the fundamentals rather than a deep dive. It assumes you are comfortable with Python and the terminal, and the local-model lessons want a reasonably capable machine. Enrollment is free; graded quizzes and a completion certificate sit behind DeepLearning.AI Pro. Few courses treat 'which agent, which model, running where' as an engineering decision with measurable tradeoffs — this one does.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Course
- Cost
- Freemium
- Duration
- ~1h 32m, self-paced (15 video lessons + graded quiz)
- Provider
- DeepLearning.AI
- Hands-on
- Yes — code/exercises
- Certificate
- Available
What You’ll Learn
- ✓How to scope tasks and context so smaller local models still produce usable code
- ✓Set up Claude Code subagents that split a job into focused, cheaper tasks
- ✓Route a coding agent through OpenRouter to swap inference providers and models freely
- ✓Configure OpenCode, an open-source coding agent, as a drop-in Claude Code replacement
- ✓Serve a local model with LM Studio and wire it into a coding agent
- ✓Measure cost, speed and token usage at each step to compare setups honestly
- ✓Reason about exactly what source code and data leave your machine in each configuration
Highlights
- •Rebuilds one identical Python app five times, so the cost comparison is genuinely controlled
- •Ends with a working fully local coding agent, not just a discussion of the idea
- •Taught by a JetBrains developer advocate and framed around sovereign AI, privacy and cost
- •Names real tools throughout — Claude Code, Haiku, OpenCode, OpenRouter, LM Studio, Gemma, Qwen
- •Ships an evaluation-metrics reading so you can repeat the same comparison on your own stack
Who It’s For
Best For
- ✓Developers whose Claude Code or coding-assistant bill is growing faster than expected
- ✓Engineers at companies that cannot send source code to a third-party API
- ✓Anyone wanting to know whether local coding agents are actually usable in 2026
- ✓Teams choosing between managed and self-hosted coding-agent stacks
Prerequisites
- •Comfortable writing Python and working in a terminal
- •A reasonably capable machine with enough memory to run local models
- •Basic familiarity with a coding agent such as Claude Code
FAQ
What is AI Coding Workflows: From Cloud to Local?
A short DeepLearning.AI course built with JetBrains and taught by Paul Everitt that walks one Python application through progressively cheaper and more private coding-agent setups: Claude Code, Claude Code with subagents, subagents on a cheaper model, the open-source OpenCode agent behind OpenRouter, and finally fully local models served by LM Studio. You finish able to argue about cost, latency and what data leaves your machine using numbers rather than opinions.
Is AI Coding Workflows: From Cloud to Local free?
AI Coding Workflows: From Cloud to Local offers free content, with paid options for certificates or premium features.
What level is AI Coding Workflows: From Cloud to Local for?
AI Coding Workflows: From Cloud to Local is aimed at a intermediate audience. Recommended background: Comfortable writing Python and working in a terminal, A reasonably capable machine with enough memory to run local models, Basic familiarity with a coding agent such as Claude Code.
How long does AI Coding Workflows: From Cloud to Local take?
Expect roughly ~1h 32m, self-paced (15 video lessons + graded quiz). Most learners work through it at their own pace.
What will I learn from AI Coding Workflows: From Cloud to Local?
You'll learn: How to scope tasks and context so smaller local models still produce usable code; Set up Claude Code subagents that split a job into focused, cheaper tasks; Route a coding agent through OpenRouter to swap inference providers and models freely; Configure OpenCode, an open-source coding agent, as a drop-in Claude Code replacement; Serve a local model with LM Studio and wire it into a coding agent; Measure cost, speed and token usage at each step to compare setups honestly; Reason about exactly what source code and data leave your machine in each configuration.
Topics
Sources
This page was written from 2 sources, 1 on domains other than deeplearning.ai.