OpenAI acquires Pierre team, integrating code viewing, review and storage capabilities into Codex

Almost at the same moment, Conductor, the parallel programming tool favored by over 100,000 coders, was also snapped up by Meta today.

Founder Charlie Holtz announced that the entire team is joining Meta Superintelligence Labs (MSL).

They will team up with former GitHub CEO Nat Friedman and Chief AI Officer Alexandr Wang to build the "next-generation interaction interface."

Writing code is only the beginning; from pulling branches and reviewing code to shipping and delivering, programmers now want every step covered.

OpenAI acquires Pierre

Codex puts "viewing, reviewing, and storing" all in its own pocket

Being spotted by OpenAI at a glance is no surprise, given the Pierre team's formidable pedigree.

Jacob Thornton, co-founder of the legendary front-end open-source framework Bootstrap, former Discord core heavyweight Amadeus, and esm.sh creator Je Xia are all part of this team.

The Pierre team even went to a street basketball court to shoot a deadpan humor short film—"no filming permit, only one shot"—to announce joining OpenAI.

Don't let the official announcement video, which looks like a ragtag production, fool you: when Tibo, the head of OpenAI Codex, reposted it, he directly likened Pierre to a computer's "RAM" and declared he would "reshape everything."

Tibo's high praise comes because, although Codex can now dispatch a dozen or more Agents to write code simultaneously, delivering the final work has become a huge headache.

Dozens of Agents hammer away in the background, and tens of thousands of lines of changes instantly flood the screen. As a result, not only can human programmers not keep up, but there is also nowhere to store such a massive number of branches.

And the three core capabilities Pierre has built up happen to fill Codex's gaps in "viewing code, reviewing code, and storing code" all at once.

Viewing code (Trees): already built into the Codex desktop version, it is specifically responsible for millisecond-level smooth expansion of massive project directories and cross-branch file location.

Reviewing code (Diffs): previously, when reviewing a PR on GitHub, even if you found that AI had written a single variable name wrong, you had to switch back to your local IDE, fix it, and push again; Pierre's Diffs directly embeds an "inline edit (Edit)" mode into the comparison view, so whichever line looks off, you can fix the code on the spot with a click.

Storing code (Code Storage): a distributed Git repository designed specifically for high-concurrency reads and writes by AI Agents, it has long been the invisible infrastructure behind popular apps such as Lovable, Bolt, and Poke. Lovable alone, a unicorn, relies on it to create over 8 million new Git repositories per week, and Poke uses it to perform 40,000 Git operations per minute to store Agent memory.

Next, this technical foundation will be fully merged into Codex; as for Pierre's original open-source projects and existing customer services, OpenAI promises everything will remain as before and will not be cut off.

AI writes Git to death

The battle burns into Microsoft GitHub's backyard

OpenAI is so eager to merge Pierre into Codex because there is actually a big embarrassment behind it—

Its own Git system was recently crashed by AI.

Just two days before announcing the purchase of Pierre, OpenAI had just released Pando, a new internally developed Git backend.

It turns out that ever since everyone at OpenAI started using Codex, engineers have been unleashing groups of Agents every day to open branches in parallel, push code, and run CI tests, and the huge internal monorepo was directly overwhelmed starting late last year.

Code couldn't be pulled or pushed, merges frequently errored out, and the reliability of OpenAI's internal Git system once fell below 90%.

Driven to desperation, OpenAI simply rewrote Git's underlying layer from scratch.

As a result, push speed became nearly 5 times faster, with nearly 1 PB of data being transferred internally every day, and even President Greg Brockman personally showed off the results.

Originally, the outside world thought this was just OpenAI's "internal self-rescue," but unexpectedly, just two days later, it turned around and bought Pierre.

With this combination punch, OpenAI has effectively charged straight into the backyard of GitHub, which is owned by major shareholder Microsoft.

Of course, OpenAI is far from the only one eyeing this juicy piece of meat.

Cursor, which launched the Origin hosting platform, Anthropic, which developed its own Scion version control, and former GitHub CEO Thomas Dohmke, who just started a company last month, are all closing in on the traditional code-hosting territory.

From editors and review consoles to the Git repository foundation, none of the big AI companies want to remain dependent on others and be merely a "plug-in" that helps people complete code.

Although GitHub, with 150 million developers, still firmly sits as the open-source home base, even GitHub's former boss has turned around and joined this army of "dismantlers."

Former GitHub boss makes his move

Intercepts Meta's 100,000-programmer tool

On the same night OpenAI bought Pierre, Zuckerberg's MSL also staged a lightning-fast talent grab.

The most noteworthy name in this acquisition is Meta's buyer, Nat Friedman.

This guy was exactly the first CEO after Microsoft bought GitHub for $7.5 billion in 2018; he personally raised the previous generation of programmer workflows.

Now at Meta, he has teamed up with Chief AI Officer Alexandr Wang and turned around to put Silicon Valley's geek circle's hottest "next-generation Agent command center" into his own pocket.

Multi-Agent isolated concurrency: you can split the dirty and tedious work among multiple Claude Code or Codex Agents, letting them work simultaneously in their own isolated workspaces without interfering with each other.

Same-screen co-op like Figma: press the shortcut "⌘⇧C" to toss the workspace to a colleague; several people can watch Agents write code on the same screen, relay instructions, and after reviewing the Diff, casually click Merge.

Even Ilya Sutskever and YC partner Aaron Epstein are early supporters.

Looking closely at the Conductor official website workspace screenshot above, you can even spot Ilya's avatar and name in the team list in the lower left corner.

Programmers' old workstations are being replaced wholesale

Today, writing code has become the cheapest step. Type two lines of prompts, and a dozen Agents can simultaneously modify code, add tests, and fix bugs.

The real bottleneck is after the code is written: how to review tens of thousands of lines of changes, how to review dozens of branches, and where to store massive repositories.

The big companies know better than anyone that models alone simply cannot lock in programmers: today Claude is handy so they switch to Claude, tomorrow GPT gets an update so they switch to GPT—switching model providers is just a mouse click away.

But once the codebase, review process, and history all live in its tools, moving out becomes difficult.

As OpenAI said in its acquisition announcement:

Every workflow left over from the old software era must be redesigned for the era in which humans and Agents write code together.

How code is written no longer matters; which desk programmers sit at to work is the real territory the big companies will fight over next.