An in-depth look at GitHub and OpenAI's Copilot, which writes 35% of its users' code, and how AI tools can make developers more efficient and democratize coding
OpenAI's new tool can autocomplete lines of programming or conjure software from a simple prompt. It could also riddle the internet with even more bugs. Tweets: @sama . Thanks: @colegarry2 Tweets: Sam Altman / @sama : An interesting case study on the direction AI tools may go: https://www.wired.com/... Thanks: @colegarry2
Context & Ripple Effects
Wired's deep dive lands between two milestones in Copilot's short life: the Microsoft-OpenAI announcement that turned a research model into a developer tool, and the October 2021 data point showing as much as 30% of new code on GitHub's network was already machine-suggested. By the time this piece ran, the tool was writing 35% of its users' code — and months later GitHub would take it out of preview at $10/month, claiming 1.2M+ developers had tried it.
The article's framing matters because it names both sides of the ledger at once: genuine efficiency gains and coding democratization on one side, and the warning that prompt-generated code could riddle the internet with more bugs on the other. That tension is exactly what the subsequent coverage plays out.
First-order effects
- Developers using Copilot now delegate over a third of their code-writing to the model, shifting their work from typing syntax to reviewing, prompting, and editing suggestions — while GitHub converts that reliance into a paid subscription.
- Teams adopting the tool inherit its error rate directly: every accepted suggestion is unreviewed third-party code entering their codebase, which is what Wired flags as the bug-amplification risk.
Second-order effects
- The quality concern hardens into evidence: a later study found programmers using Copilot and similar tools produce less secure code than those who don't — while believing their code has no safety issues — forcing buyers to weigh productivity against an invisible security tax.
- Copilot's traction pressures every other developer-tools vendor to ship an AI assistant or cede the editor, turning code completion from differentiator into table stakes across IDEs and hosting platforms.
Third-order effects
- If AI-written code keeps climbing as a share of the world's software, the industry's bottleneck moves from writing code to assuring it — review, testing, and security tooling become the growth layer, and 'democratized' coding means more non-experts shipping code they cannot fully vet.
- The trajectory points past autocomplete entirely: GitHub's later move to an AI agent that fixes bugs and adds features autonomously suggests the assistant becomes the operator, restructuring engineering teams around supervision of machine-written software.
The trend: Software development is shifting from human-authored to AI-assisted and eventually AI-agent-authored code, with quality-assurance and security practices lagging behind adoption.