Lesson 1 / 29

From Autocomplete to Agents

See the three generations of AI coding help and what changed.

Suggest, converse, act

AI coding tools arrived in three overlapping waves. Autocomplete predicts the next few tokens or lines inside your editor: fast, low-risk, you stay in control. Chat assistants answer questions and draft code in a conversation, but you copy results between the chat and your project. Coding agents are given a goal ("fix this failing test", "add pagination to the orders list") and then act: they search the repository, open and edit files, run the build and tests, read the output and try again, often over dozens of steps. The shift is from suggesting text to taking actions and checking results, which makes agents far more capable and also far more consequential when they are wrong. Product names and features in this area change monthly, so this course teaches the ideas that stay constant; check each tool's current documentation for specifics.

A model in a loop with tools

A coding agent reads your code, edits files, runs commands and tests, observes the results and repeats until the task is done or it stops.

Four parts: model, tools, context, loop.
Figure 1.1 — Model, tools, context and loop.

Three generations at a glance

What each generation does and who is in control.

Generation        What it does                               You control          Typical risk
autocomplete      predicts the next tokens/lines in-editor   every keystroke      subtly wrong snippets
chat assistant    answers, drafts code in a conversation      what you paste       outdated or invented APIs
coding agent      plans, edits files, runs commands/tests     goal, permissions,   wrong changes at scale, side effects,
                  and iterates on the results                 review of the diff   runaway cost, prompt injection

Match the tool to the task

Use autocomplete for typing speed, chat for questions, and an agent for multi-file tasks with a clear check.

Quick check: What most clearly distinguishes a coding agent from a chat assistant?

  • It only works offline
  • It never makes mistakes
  • It takes actions (edits files, runs tests) and reacts to the results
  • It replaces version control
Answer

It takes actions (edits files, runs tests) and reacts to the results — Acting and observing in a loop is the defining feature.