GitHub Next

We are a team of researchers and engineers at @GitHub who investigate the future of software development.

[GitHub Next @GitHubNext]

A new report from GitHub Next!

The Impact of Automated Repository Maintenance Assistance

https://t.co/XrkecLmc7u

What happens when a proactive AI repository agent is deployed across 13 open source repositories? Hundreds of issues closed, median ~8x increase in issue closurea nd PR merge velocity - transforming largely dormant projects into actively maintained ones.

The single most important factor? The rate at which human maintainers decide to act.

#ContinuousAI #RepoAssist #GitHubNext

[GitHub Next @GitHubNext]

@PromptSlinger @zeke @idangazit @Mappletons gave a demo of it at AIE London, the visuals are a bit out of date but the core concept is there, in lovely essay form:

https://t.co/pThxBJILrE

[GitHub Next @GitHubNext]

@PromptSlinger @zeke @idangazit

[GitHub Next @GitHubNext]

@ilbert_luca @shadcn @Mappletons There wasn't a write up we removed, more that we have been heads down on building and we sucked at spending time on the storytelling.

The best resource right now is @Mappletons 's talk but that's on her site: https://t.co/pThxBJILrE

[GitHub Next @GitHubNext]

@ilbert_luca @shadcn @Mappletons Not shut down! We're still working very hard on it

[Mario Rodriguez @mariorod1]

The aim of agentic workflows is to make the repo and PDLC a living organism. Think a garden. Alive, evolving, and always improving.

We have a wide catalog available and we will continue to expand them.

https://t.co/uJW4Y63nXw

[Oren Melamed @OrenMe]

This is huge!
We’ve been partnering with @GitHubNext team and this is a real thing!
I know we talk a lot about coworks and Hermes etc but real engineering happens in pipelines and taking advantage of GitHub actions as pipeline runtime plus having the ability to run @GitHubCopilot CLI agent harness there, in a secure manner !!!, is the way to scale where engineering goes to die - endless manual tasks can be agentified

And don’t get me wrong - as much as I say it is good it has been a struggle to run it in corporate constrained environment and there is work that can be done e there to improve

Also cost management is hard and there are no clear controls to stop spend and u need to build your own

But in good timing they also released this https://t.co/HuPGkJ2DAl and believe me that this is a game changer cause the previous way on minting tokens was impossible

[GitHub Next @GitHubNext]

OMG the one-liner we wish we'd said

[Hassan @buildwithhassan]

@github github actions but the action is an agent now

[GitHub Changelog @GHchangelog]

Agentic workflows can now use the built-in GITHUB_TOKEN instead of personal access tokens, reducing security risks and simplifying automation.

https://t.co/2X3mo82jJL

[GitHub @github]

GitHub Agentic Workflows are now in public preview.

Intelligent automations for GitHub, with strong guardrails, observability, and cost controls. Start automating today 👇
https://t.co/Z3Zcx48om8

[Mastra @mastra]

Python trains, TypeScript ships.

Tickets just dropped for TypeScript AI Conference: London → https://t.co/hwSgNVVAR4

Featuring speakers from @GitHubNext, @elastic, @sentry, @arizeai, @CopilotKit, @factorialapp, and others.

July 23 | London & Virtual | Builders only

[GitHub Next @GitHubNext]

Agents Need Boundaries Too!

Read about this GitHub Next exploration of reusable harm gates, to keep the agents in the channel to success, even when subverted by attackers.

https://t.co/T7o69UmTmm

#GitHubNext #AgenticAI

[GitHub Next @GitHubNext]

Read about "Agent Functions", a new agentic design pattern for structuring agents, from GitHub Next

https://t.co/Fa7T1gd8BS

Prompts are programs. You wouldn’t write a complex program completely from scratch, in a big, soupy loop without subroutines, and then write it again the next time you wanted to run it. Why let your agent work that way?

[GitHub Next @GitHubNext]

At GitHub Next, the Autoloop agent is porting Pandas to Typescript

Read about this experiment in simple, declarative goal-directed repository automation!

https://t.co/0dqXVnEqQp

[GitHub Next @GitHubNext]

Ace👀

[Terkel @terkelg]

loading states that bleed outside the window

[GitHub Next @GitHubNext]

We have a draft of new report from GitHub Next! 🚀

https://t.co/2oEAVO7OCD

The Impact of Automated Repository Maintenance Assistance

- We analyze the impact of 🌈 Repo Assist, a proactive AI repository agent implemented as a GitHub Agentic Workflow, across 13 open source repositories in early 2026.

- The agent reduced open issue counts in every repository - 578 issues total.

- The repositories, which were largely dormant, achieved median velocity increases of 8× for issue closure and 10× for PR merges, transforming largely dormant projects into actively maintained ones.

- We model repositories as human-agent software factories - the factory’s throughput is gated by human decision-making.

Read the draft in our repository https://t.co/2oEAVO7OCD

[GitHub Next @GitHubNext]

Thanks for the feedback. We will add notes along these lines:

* The rate of "/repo-assist" human guided agentic intervention is mentioned in the report.

* There was little other review override - some but not much.

* Reopen rate was essentially zero.

* Learning memory repo norms is not part of the spec of repo-assist, it could be added through memory.

[GitHub Next @GitHubNext]

We have a draft of new report from GitHub Next! 🚀

https://t.co/2oEAVO7OCD

The Impact of Automated Repository Maintenance Assistance

- We analyze the impact of 🌈 Repo Assist, a proactive AI repository agent implemented as a GitHub Agentic Workflow, across 13 open source repositories in early 2026.

- The agent reduced open issue counts in every repository - 578 issues total.

- The repositories, which were largely dormant, achieved median velocity increases of 8× for issue closure and 10× for PR merges, transforming largely dormant projects into actively maintained ones.

- We model repositories as human-agent software factories - the factory’s throughput is gated by human decision-making.

Read the draft in our repository https://t.co/2oEAVO7OCD