A developer types def is_valid_email(address): and grey text appears with a complete function body. Pressing Tab accepts it; typing anything else dismisses it. Editors have offered completion for decades, but older tools listed names the compiler already knew. A language model instead predicts whole lines from the code around the cursor, including logic nobody has written yet.
Completion was the first widely used AI coding product. GitHub Copilot launched on a production version of OpenAI’s Codex model, and the Codex paper’s HumanEval benchmark became a standard way to measure code generation (see pass@k). Later tools also predict the next edit elsewhere in the file, not just the next line.
Guests on the show treat completion as the starting point of a progression. In episode 40 at 6:28, Graphite’s Greg Foster remembers AI tab complete filling the rest of a line, or a few lines, and people paying for it. In episode 23 at 23:11, Poolside’s Eiso Kant traces the path from code completion to chat to early agents. In episode 10 at 10:08, Block’s Rizel Scarlett recalls early Copilot users expecting an agent and finding something closer to tab completion. Episodes 23 and 40 were sponsored by Galileo, which Conor mentions in episode 23 at 19:01 and plugs in episode 40 at 2:05.
The developer stays in control line by line, which makes completion lower-risk than handing a task to a coding agent. A plausible suggestion can still be wrong or insecure, so read what you accept and keep tests in place. AI code review and vibe coding sit further along the same spectrum.
Hear it from the guest
“Now I could use AI and it could predict a little bit better and it could fill up like the rest of the line”
“We came from code completion to chat to now early agentic”
Quotes lightly edited to remove filler words.
Sources
- Chen et al.: Evaluating Large Language Models Trained on Code (2021) — Introduces Codex, a GPT model fine-tuned on public GitHub code, notes that a distinct production version powers GitHub Copilot, and releases the HumanEval benchmark; checked October 24, 2026.
- GitHub Docs: GitHub Copilot in IDEs — Lists inline and next edit suggestions alongside chat and agent features, and describes code referencing for suggestions that match public code; checked October 24, 2026.
