xbin Try it free

The first exponential era workspace

Software has entered its exponential era.

Agents now make software faster than companies can run it.

Try it free for an hour Install

Sheets of yellow, green, magenta and cobalt glass hanging in a concrete hall, laying coloured light across the floor.

Seventy years

Seventy years of software. Then the curve bent.

For seven decades, software grew as fast as people could write it. Then the first large language models learned to write code too.

Software made per year: new repositories created on GitHub per year

New repositories created on GitHub per year, 2013 to 2026: from 5.5M in 2013 to 121M in the year to August 2025, then 288M in August 2026 (annualised August rate); marked: First large language models, June 2021.
Show the numbers
New repositories created on GitHub per year (repositories/year). Repositories created in the Octoverse year (12 months ending ~Sep 30; Aug 31 for 2025). 2018 is derived (≈1/3 of 96M). 2026 is an annualised monthly rate, not a full-year total.
DateNew repositories created on GitHub per yearSource
2013-12-235,500,000github.blog · "over 5.5M repositories ... created this year alone"
2018-09-3032,000,000github.blog · derived: "one third of all repositories were created in the last 12 months" x 96M
2019-09-3044,000,000github.blog
2020-09-3060,000,000octoverse.github.com
2021-09-3061,000,000octoverse.github.com
2022-09-3085,700,000octoverse.github.com
2023-09-3098,000,000altusintel.com
2024-09-30108,000,000github.blog
2025-08-31121,000,000github.blog
2026-08-15288,000,000github.blog · annualised: "new repositories per month rising to about 24M" x 12 (GitHub, Aug 2026)
  • 121M

    new repositories on GitHub in Sep 2024–Aug 2025

Linear scale. Sources: GitHub, its Octoverse reports of 2018 to 2025 and its blog posts of December 2013 and August 2026, and for 2023 a summary of the Octoverse report (altusintel.com). The dotted end is a rate, not a year: about 24M new repositories in August 2026, times twelve.

Today

The old way breaks.

  • Every new tool needs a team to run it.
  • Every change waits in a queue.
  • Every agent doing a person’s job is one more thing to watch.

Agents make software faster than any of that can keep up. Companies need a new way to build what they run on.

The idea

A company is a set of systems that get things made.

Agents should build those systems.

People and agents build them, and no longer operate them.

What xbin is

xbin is a workspace where people and AI agents build the systems a company runs on, and where those systems run.

  • Describe a change to an agent, or make it yourself. Save, and it is live.
  • Every app and every agent works in its own sandbox and reaches only what someone has granted.
  • A person can have their own copy of an app. Other people cannot read it, and admins do not open it.
See how it works

Measured

It works today.

  • page saved to page on screen under0.35 s

    On a 96-core workstation, a saved page is on screen in under 0.35 s, including a deliberate 0.3 s settle.

  • reads failed in 60 live updates 0

    On a 96-core workstation, 0 of 5.39 million reads failed across 60 live updates under load, at about 30,000 requests a second.

  • a person’s own copy, first request under0.3 s

    On a 96-core workstation, a person’s own copy of a small app starts on their first request in under 0.3 s.

Measured on a big workstation: an AMD Ryzen Threadripper PRO 7995WX with 96 cores and 755 GiB of RAM, running xbin v0.3.67 on 3 October 2026 alongside other work. Each figure is the slowest sample or the 90th percentile, rounded up. A smaller machine was not measured. Writes that land at the exact moment of an update can fail: 15 of 2.34 million did.

Upgrade to exponential era software

Try it free for an hour

ssh xbin@vcpu.sh

All you need is an ssh client.

Install

curl -fsSL https://xbin.dev/install.sh | sh

The installer prints its plan and asks before it changes anything.

Build the systems your company runs on.

People and agents build them in xbin. The systems do the work.

Try it free for an hour Read the docs ↗ Source on GitHub ↗