AIght_
ToolsLearnFieldsUniverseSignalHumanAbout
Take the quiz
← All fields

Field guide

Software Engineering

AI is moving from autocomplete to autonomous agents, transforming coding into high-level system orchestration where the 'how' is automated so you can focus on the 'what'.

Medium
Start your path →See your personal disruption score in 2 minutes

What's changing

01

AI-native IDEs and agents like Cursor and Windsurf understand entire codebases, allowing engineers to refactor, debug, and implement features across multiple files with natural language commands.

02

Large language models specialized for code generate idiomatic boilerplate, unit tests, and documentation, reducing routine 'grunt work' by up to 80% for senior developers.

03

Autonomous coding agents and 'software engineers' like Devin or OpenDevin handle end-to-end tasks from bug fixing to deployment, shifting the human role toward architecture and review.

Copilot completes about 30% of code by lines in some codebases. The other 70% is where the real disagreements live.

A path through the universe

How to actually learn AI for Software Engineering.

Two tracks. Pick your depth. The left one gets you fluent for conversations and tool choices. The right one is what you read when you actually want to know how it works.

Intuitions

No math required.

  1. 01Prompt EngineeringThe craft of talking to a model that will take you exactly as literally as it decides to7 min read
  2. 02Function CallingThe JSON-shaped API that turned chat models into clients of the real world.6 min read
  3. 03Chain-of-ThoughtWhen 'think step by step' actually earns its keep — and when it's just expensive theater.6 min read
  4. 04Structured OutputForcing the model to fill in a shape — and why it's harder than it looks.5 min read
  5. 05Context WindowsWhat the model can see right now — and why the edges matter6 min read
  6. 06In-Context LearningHow models 'learn' from examples in the prompt — without changing a single weight.6 min read

Goes deeper

Under the hood.

  1. 01Model Context ProtocolThe open standard that lets AI models talk to your tools without a custom integration per model7 min read
  2. 02Retrieval-Augmented GenerationHow AI learned to look things up before opening its mouth8 min read
  3. 03AI AgentsWhen AI stops answering and starts doing — and then, very often, hits a wall9 min read
  4. 04KV CacheWhy long conversations are cheaper than they look — and the reason your API bill behaves the way it does.5 min read
  5. 05AttentionThe single mechanism behind every model since 2017 — and the one that quietly burns most of the compute.8 min read
  6. 06Fine-TuningTeaching a model new habits, not new knowledge8 min read

AI impact spectrum

Automated

  • Routine tasks
  • Data processing
  • Standard reporting

Augmented

  • Analysis
  • Decision support
  • Research

Growing

  • Judgment
  • Relationships
  • Creative work

Roles at risk

Junior web developer (boilerplate/UI)

Routine QA / Manual tester

Documentation writer

Maintenance engineer

Roles growing

AI system architect

AI/ML engineer

Engineering team orchestrator

Security and compliance auditor

Boilerplate-heavy roles (CRUD APIs, glue code) compress fastest. Systems design, debugging, and review compress slowest.

What to actually do

Software engineers must shift their focus from writing syntax to architecting systems and reviewing AI-generated code, using agentic IDEs like Cursor for every feature implementation to maximize velocity. Dedicate time to building custom Model Context Protocol (MCP) servers to give AI agents access to internal tools and documentation, implement rigorous automated testing to validate AI outputs, and master the art of system design over manual implementation. Treat AI as a highly competent junior dev that needs clear direction and constant review, ensuring you remain the ultimate authority on security, performance, and long-term maintainability.

Spend a week with Cursor or Aider on a codebase you know. The places where you slow down to argue with the model are where you're learning the most.

Sources

  1. [1]Chen et al., Evaluating Large Language Models Trained on Code (Codex paper, 2021)
  2. [2]GitHub, Quantifying GitHub Copilot's impact on developer productivity (2022)
Medium

Engineers are early adopters, but the shift from 'doing' to 'orchestrating' requires a significant mental model change and mastery of new agentic tooling.

Personalize this

How disrupted are you, really?

Three questions. An honest score tailored to your specific role.

Take the quiz →

Tools to know

Cursor

AI-native code editor that indexes your whole project for context-aware coding and refactoring

Windsurf (Codeium)

Agentic IDE that can autonomously navigate codebases and perform multi-step engineering tasks

GitHub Copilot

The industry-standard AI pair programmer that provides real-time autocomplete and chat assistance

Concepts to understand

Agentic workflows in software engineeringModel Context Protocol (MCP) for tool integrationRetrieval-augmented generation for codebases

Get your personal disruption score

Based on your specific role within Software Engineering

Run AI impact quiz →Explore other fields