Then: a suggestion inside the editor
GitHub introduced Copilot’s technical preview in June 2021 around suggestions for whole lines and functions, using the surrounding code as context. Its launch material showed an AI pair programmer inside the editing process. A year later, GitHub reported more than 1.2 million developers had joined the preview. That is evidence of adoption, not the view count of a viral post.
Now: a repeatable job with a review point
An August 2026 OpenAI account describes using Codex with Runme notebooks for recurring evaluation work. The notebook supplies a goal, a plan to review and a place to document commands and results. The unit of work has expanded from a suggested code fragment to a process with tools, records and checkpoints.
These are two documented examples from different years, not a controlled performance comparison. They do not prove that every modern coding agent can complete every project, or that earlier tools had no broader uses.
What changed for the person using it
The useful question has moved from whether you accept a suggestion to how you define and verify a task. A good handoff names the desired result, the evidence needed to call it done and the decisions that remain yours.
For your own recurring work, try writing that brief before picking a model. An impressive output is easier to judge when you know what successful completion means. The historical images show the original products in their respective eras, without presenting either as a launch today.
Source published 2021-06-29. Coverage is based on the maker’s announcement and demonstration.