AI Developer Roadmap
Start with Python, APIs and data validation. Connect a language model, handle its responses, and add retrieval or tool use when the application needs it.
Start with Python, APIs and data validation. Connect a language model, handle its responses, and add retrieval or tool use when the application needs it. Develop one application through eight stages, test its behaviour, and deploy it on AWS. Use Claude Code or GitHub Copilot to assist development while you review and test the code.
Sources and methodology · This roadmap is reviewed when production practices, tools or platform patterns materially change.
Stages
8
Last reviewed
18 October 2026
Before You Start
Before You Start
You can follow this roadmap with a background in Java, JavaScript, C#, Python or another programming language. The examples use Python, but your experience with APIs, databases, debugging and application design remains useful.
If you already build software, focus first on Python syntax, virtual environments, type hints, JSON and HTTP requests. If you are new to programming, practise functions, data structures, exceptions, file handling and basic Git before beginning the application.
You do not need to train a model before building your first AI application. You do need to understand the model's inputs, outputs, limitations and the checks your application must enforce. Use the Python for AI Development guide for language foundations and a response-validation example.
Code with AI throughout the roadmap
Use Claude Code or GitHub Copilot for small implementation tasks, test drafts and explanations of unfamiliar code. Give the assistant a clear task and relevant repository context, inspect its changes, and run checks before accepting them. Vibe coding can help you explore an idea; progressing through this roadmap requires understanding, debugging and explaining the resulting application.
Tool documentation: Claude Code · GitHub Copilot
Eight stages
Eight Stages from Python to an AI Application
Use the stages to develop a documentation and account-support application. Begin with a small API, then add model responses, retrieval and a restricted account-status tool. Keep a record of expected behaviour from your first model integration and extend those checks as the application grows.
Stage 1: Build the Python and API Foundation
Learn the Python needed for application code: functions, exceptions, type hints, JSON, dependencies and environment configuration. Use FastAPI for endpoints, request validation and a small persistence layer.
A model bolted onto an unstable application produces failures that are hard to attribute.
- What you learn
- FastAPI endpoints and request validation.
- Persistence for document and account records.
- Type hints, JSON and environment configuration.
- API tests: valid, invalid and missing records.
- What you should build
- Create an API that stores document records and fictional account records. Add tests for valid requests, invalid payloads and records that do not exist.
- Ready when
- Someone can run the application from your setup instructions. Records persist correctly, invalid requests receive clear errors, and the API tests pass.
- Common mistake
- Skipping tests and validation because the model will "handle it".
- Related resources
- FastAPI documentation — Endpoints, request validation and testing patterns
Portfolio
Build a Portfolio Application on AWS
Bring the stages together in a documentation and account-support application. It should answer questions from a defined document collection, show supporting sources and retrieve account status through a restricted tool. Use fictional account data for the demonstration.
Your portfolio should show how the application behaves when information is missing, a dependency fails or a request is outside the user's permissions.
Deliverables
- Source code and reproducible setup instructions.
- An explanation of the request flow, data boundaries and deployment.
- Evaluation cases, a baseline and an example of a measured improvement.
- Examples of unsupported questions, invalid tool requests and interrupted responses.
- AWS configuration notes, recovery steps and resource cleanup instructions.
- A short demonstration and an honest account of remaining limitations.
Present the project through the problem, implementation decisions and evidence from testing. A deployed URL alone does not explain whether the application behaves reliably. Explore additional AI developer project briefs if you want to practise another application pattern.
FAQ
AI Developer Roadmap FAQs
Direct answers for software developers following this roadmap.
Can I follow this roadmap if I use Java or JavaScript?
Yes. Your experience with application design, APIs, databases and debugging transfers. Learn the Python syntax and tooling needed for the examples, then use the stage checks to identify any gaps in your application-development skills.
Do I need machine learning or advanced mathematics first?
You can begin API-based AI application development without advanced mathematics or model-training experience. Learn the model concepts needed to reason about context, outputs, evaluation and limitations. Training or adapting models requires additional foundations.
Should I learn RAG before AI agents?
Start with a working API and a model integration. Add RAG when the application needs answers from external documents. Add tools or agent workflows when it needs to interact with systems. These capabilities solve different problems, so neither is mandatory for every application.
Can I use Claude Code or GitHub Copilot while learning?
Yes. Use them to assist implementation, understand unfamiliar code and draft tests. Review the changes and verify behaviour yourself. You should be able to explain the code and diagnose failures without relying on the assistant's explanation alone.
How long does this roadmap take?
The time depends on your existing programming experience, weekly availability and project scope. Use the stage checks to assess progress. A developer comfortable with APIs and testing can move through the foundations faster than someone learning those skills for the first time.
How is an AI developer roadmap different from an AI engineer roadmap?
Job titles overlap across organisations. This roadmap concentrates on building applications with models, retrieval and tools. An AI engineering path may also cover model training, adaptation, inference infrastructure and broader ML systems. Compare the AI developer and AI engineer paths and check the responsibilities of the specific role or programme.
Continue
Choose Your Next Learning Step
You can follow this roadmap independently using the stage checks and project deliverables. If you want live teaching, guided projects and feedback on your code and architecture, explore the School of Core AI AI Developer Course. The course includes AI-assisted development and an AWS-deployed capstone.
Continue with the Agentic AI roadmap for deeper work on agent state, orchestration and tool control.
Continue with the LLMOps roadmap for model serving, evaluation infrastructure and operating LLM systems.