Copilot didn’t guess. It had been silently indexing her entire repo for the last 24 hours. It replied: Based on migrations/2024-03-12_init.sql and models/legacy/UserEvent.js , the table has columns user_id (uuid), action (varchar), and created_at (timestamp). Note: action has a typo: ‘subscrioption’. Do you want me to fix that in the new migration? Maya blinked. “Yes. And also…” she started typing, but Copilot interrupted (politely): “I see you also have a user_preferences table. The new time-series DB (QuestDB) doesn’t support JSONB. Would you like me to flatten those fields into columns?” This was the update. It had learned her team’s naming conventions, common bugs, and even her lead’s preference for explicit error handling. 3. The “Safety Catch” (Policy Enforcement) At 3 PM, Maya got reckless. She asked Copilot in plain English: Write a cron job that deletes inactive users.
Copilot replied: “You’re welcome, Maya. I’ve also backported your retry logic to the release/2025.12 branch. See you tomorrow.” The November 28, 2025 updates didn’t just add features. They turned GitHub Copilot from a code completion tool into an autonomous, context-aware, safe engineering partner — one that finally understood that the best code is the code you don’t have to write. github copilot updates november 28 2025
A small terminal panel opened. Copilot replied in a calm, text-to-speech voice: “You’re mutating jobStatus inside a Promise.all without a lock. On line 47, two jobs complete at the same millisecond and overwrite each other’s success flag. Recommend using p-limit or a Redis atomic counter.” It then opened a terminal tab , typed the fix command, and ran the tests. All green. Maya closed her laptop at 5:02 PM. She had shipped all three tickets. Normally, that was three days of work. Copilot didn’t guess