GitHub dropped its 2026 Octoverse report at Universe yesterday, and the numbers are wild. Merged pull requests tripled in less than a year to 220 million per month. Commits more than tripled to 7.38 billion per month. Most of that increase happened in the second half of this year.

This isn’t speculation about what AI might do to software development. It’s a measurement of what already happened.

The agent effect

The timing tells the story. Simon Willison has been tracking what he calls “inflection points” for coding agents. He pegged November 2025 as the moment when Claude Opus 4.5 and GPT-5.1 became reliable enough for daily use. The GitHub data backs that up. Activity was growing steadily, then it shot up.

Think about what tripling commits in a year actually means. Either developers suddenly got three times faster at typing, or something fundamental changed about how code gets written. We know which one it is.

What this looks like in practice

The Octoverse numbers are aggregate data, but we’re seeing what this looks like on the ground. Willison also wrote about reaching another inflection point in October 2026, this time for what he calls “vibe coding” games. People are building actual playable games by describing what they want and iterating with agents.

He points to examples like Tree Appreciator, Emberwick Battle Drones, and Pumpkin Farmer (also called Quarter Sun Cove). These aren’t tech demos. They’re games people made without writing most of the code themselves.

Cursor shipped an Agent SDK this week too. The pattern is clear: agents aren’t experimental anymore. They’re infrastructure.

The Stack Overflow signal

Here’s another data point. Stack Overflow now gets fewer than 1,000 questions per month, according to a query against their public data. That’s not a typo. The site that defined how developers got help for two decades is nearly silent.

You can read that a couple ways. Maybe developers aren’t getting stuck anymore. Maybe they’re asking agents instead. Maybe both. Either way, it’s a massive behavioral shift.

Not everything benefits

Amazon engineering leadership put out a note this week (via Terk Terkowitz) that’s worth reading. The key line: “If your work does not benefit from current AI tooling, that is a finding, not a failing.”

That’s good advice. The GitHub numbers show a real change, but they don’t show that change helping everywhere equally. Documenting where agents don’t help yet is just as important as celebrating where they do.

The Octoverse data gives us a baseline. We can now measure what “the age of agentic engineering” actually looks like in activity metrics. Seven billion commits per month is the new normal. The question isn’t whether agents changed how we build software. It’s what we build now that we couldn’t before.