# AI coding assistants, from autocomplete to agents

URL: https://ai-pulse.aglaganova.com/blog/ai-coding-assistants-from-autocomplete-to-agents
Published: 2026-10-09
Author: Roman Aglaganov, Full-stack developer
Category: Tools
Region: Global
Tags: coding, agents

> In 2022, GitHub found developers finished a task 55% faster with Copilot. Three years later, coding assistants do not just finish your line; they open the files, run the tests and propose the fix.

![Typographic cover: 55% faster, in GitHub’s 2022 Copilot experiment](https://ai-pulse.aglaganova.com/api/media/file/coding-1.png)

Software developers were among the first professionals to get an AI colleague, and probably the first to argue with one about tabs versus spaces. GitHub Copilot, built with OpenAI, launched as a preview in 2021 and became generally available in June 2022. It started as very clever autocomplete: write a comment or the start of a function, and it suggests the rest.

## The 55% experiment

In 2022, GitHub ran a controlled experiment with 95 professional developers, asking them to write a small web server in JavaScript. The group using Copilot finished, on average, 55% faster than the group without it, and slightly more of them finished at all. One task is not a whole career, but it was the first hard number in a debate that had mostly run on vibes.

## From suggestions to agents

The next wave does more than suggest. Agentic coding tools, including Anthropic’s Claude Code, OpenAI’s Codex, Cursor and Copilot’s own agent mode, can read a whole codebase, edit several files, run the test suite and try again when something fails. You describe the bug; the tool investigates, like a junior developer who never needs coffee and occasionally deletes the wrong folder.

## What has not changed

**Review still matters:** AI-written code can look plausible and still be subtly wrong or insecure, so teams treat it like any colleague’s pull request.

**Understanding still matters:** the developers who get the most out of these tools are the ones who can tell when the suggestion is nonsense.

The punchline: AI did not replace programmers. It replaced a lot of typing, and turned many programmers into reviewers, which some of them say is the harder job.

## Sources

- [GitHub: Quantifying Copilot’s impact on developer productivity](https://github.blog/news-insights/research/research-quantifying-github-copilots-impact-on-developer-productivity-and-happiness/)