Chaos and Order

💡 왼쪽 원문을 읽으면서 오른쪽에 따라 써보세요. Tab 키로 힌트를 받을 수 있습니다.

Introduction — 776 Points, and Still No Product
What Was Disclosed, and What Wasn't
Why a Terminal Multiplexer — the Asset Called libghostty
What Ghostty Built, and Why It Isn't a Company
From HashiCorp to IBM — What the Track Record Actually Tells You
Testing the "Going Back to Low-Level Work" Narrative
The Skepticism Raised in the HN Comments
Conclusion — Saying Thin When an Announcement Is Thin
References

Introduction — 776 Points, and Still No Product

On July 29, 2026, Mitchell Hashimoto unveiled a new company, Superlogical, and posted a short piece on his own blog. The same day, the company site landed on Hacker News and picked up 776 points and 452 comments, while his blog post drew 150 points in a separate thread.

A reaction this size isn't an evaluation of the product itself. It's because there isn't a product yet. The site is taking beta signups, and what's actually been disclosed is only the problem statement, the team makeup, and the shape of the first product. The 776 points are a reaction to the fact that the person who built HashiCorp and Ghostty is starting again.

So this post tries to keep two things separate. First, what was actually announced. In an early-stage announcement, information is usually thin, and that thinness is itself information. Second, what this choice suggests about where value is moving in infrastructure tooling. Here I'll draw a clear line between what's suggestive and what's speculation. Up front: the range of things this post can treat as confirmed fact is narrower than you'd expect.

What Was Disclosed, and What Wasn't

Let's start with the facts. Hashimoto's own blog post and the company site disclose different amounts. The blog post has neither a co-founder's name nor any investment information; the team and investor list are on the company site.

Item	Status
Company name and mission	Disclosed. The site's copy amounts to "building a multiplexer for every kind of work"
First product	Disclosed. A terminal multiplexer. Web and native macOS/iOS clients, long-lived sessions, and built-in session sharing
Team	Disclosed. Hashimoto plus three others
Investors	Disclosed. Notable Capital, Amplify Partners, and a number of angels
Funding amount	Undisclosed. Only the existence of a seed round is confirmed
Valuation	Undisclosed
Launch timeline	Undisclosed. Only a beta waitlist exists
Revenue model	Undisclosed
Relationship to Ghostty	Disclosed. Uses libghostty as an MIT-licensed component; Ghostty remains a nonprofit
Press coverage	None as of this writing

The four-person team looks like this. Hashimoto (HashiCorp co-founder; Vagrant, Terraform, Vault, Ghostty), Jack Pearkes (HashiCorp's first employee, former VP of Engineering and VP of R&D), Alasdair Monk (Head of Experience at Poolside, former VP of Design at Vercel, with a HashiCorp and Heroku background), and Hector Simpson (a designer at Poolside with a background at Heroku, HashiCorp, Clearbit, and Vercel). Two from HashiCorp, two from Poolside.

On the investor side, what's confirmed is that Notable Capital's Glenn Solomon said he led the seed round, and Amplify Partners disclosed its own investment. Solomon is also the person who led HashiCorp's Series A back in 2014. The angel list includes Patrick Collison, Tobias Lütke, Guillermo Rauch, Aaron Levie, and Armon Dadgar, among others. No one has disclosed an amount anywhere. If you come across a number, check the source.

The most information-dense sentence in Hashimoto's own post is his explanation of why he started the company: he'd found a problem worth spending years on, its scope required a team, and there were people he wanted to build it with. And he draws an explicit line around Ghostty — Ghostty stays a nonprofit; its mission, governance, license, technical goals, and roadmap don't change.

Why a Terminal Multiplexer — the Asset Called libghostty

A terminal multiplexer isn't a new category. screen dates to 1987, tmux to 2007, and there are several more recent implementations like zellij. It's no surprise HN comments immediately reacted with "another tmux?"

What Superlogical describes as the problem is a bit different. The items the site lists are local development, remote access, coding agents, background jobs, production applications, live debugging, sandboxes, shared terminals, and incident response. In other words, the diagnosis is that the number of places work happens has grown, but the tooling still treats them as separate worlds. On top of that comes a web client, an iOS client, and session sharing.

The technically interesting part is the raw material. libghostty is Ghostty's terminal-handling core split out as a library, released under the MIT license. Hashimoto states explicitly that Superlogical uses this component under the same terms as anyone else.

Here's why that matters — terminal emulation is the kind of swamp only someone who's built one truly understands. Escape sequence compatibility, Unicode width calculations, ligatures and font fallback, scrollback memory management, GPU rendering. Building this layer from scratch takes years. The fact that someone who's already built that layer once is starting out with it as a ready-made library is the real asset behind this company. Put another way, the barrier to entry isn't in the component — it's in what sits above it: session synchronization, remote connectivity, a permission model, and giving people a reason to beat the muscle memory they've already invested in their tmux config.

What Ghostty Built, and Why It Isn't a Company

To read Superlogical, you have to look at Ghostty's last three years. It's a GPU-accelerated terminal written in Zig; 1.0 shipped on December 26, 2024. The latest stable release as of this writing is 1.3.1 from March 13, 2026, and as observed on July 31, 2026, the GitHub repository sits at 59,000 stars and 3,200 forks.

What's worth noting isn't the release numbers so much as three governance decisions.

February 7, 2025 — the subsystem-maintainer model. The project eliminated the notion of a "core maintainer" who oversees the whole thing, and instead gave autonomy to eight people across specific areas — terminal emulation, Linux and GTK, shell integration, fonts and rendering. What's interesting is that the model explicitly states there is no obligation to review PRs and issues. It's a design that recognizes the main cause of open-source burnout isn't authority — it's a sense of obligation.

December 3, 2025 — the nonprofit conversion. Instead of founding an independent foundation, the project entered a fiscal-sponsorship arrangement under Hack Club, a 501(c)(3). Intellectual property transfers to Hack Club, and contributors retain copyright under the MIT license. In the announcement, Hashimoto states he remains the project lead and final decision-maker, while pinning down, in the same breath, that no funds go to him personally, and none are used in a way that benefits him. His family separately donated $150,000 to the project.

April 28, 2026 — the decision to leave GitHub. Citing recurring GitHub Actions and platform outages, the project decided to move its repository. At the time of the announcement the destination was undecided; a read-only mirror stays on GitHub, and Hashimoto's personal repositories are left as-is. This is still an open matter as of this writing.

Stack the three decisions together and the direction is clear. Ghostty was deliberately designed so that it couldn't become a company. The trademark and IP sit with the nonprofit, funds don't flow to the founder, and authority is distributed by area. So when Superlogical's announcement says "Ghostty doesn't change," it isn't just a pleasantry — it's confirming something that was already structurally locked in.

I think this contrast is itself the most interesting part of this case. Prior cases have shown plenty of what kind of conflict arises when an open-source project and a company come out of the same person. Here, those points of conflict were resolved before the company was ever founded.

From HashiCorp to IBM — What the Track Record Actually Tells You

In chronological order:

December 14, 2023 — Hashimoto announces on the company blog that he's leaving HashiCorp. He'd stepped down as CEO in 2016 and stepped away from day-to-day management in 2021, so this wasn't his first retreat — it was the final one. In the post he writes that his interests as an engineer lay beyond infrastructure, and that he'd always known he would eventually move to a different challenge.
April 24, 2024 — IBM announces its acquisition of HashiCorp.
February 27, 2025 — the acquisition closes. $6.4 billion by enterprise value, $35 per share in cash.

I should note what I could not confirm here. I found no record of Hashimoto commenting publicly on the IBM acquisition. Nor did I find a record of him commenting publicly on Terraform's BUSL license change in August 2023. I searched his HN account for that period and came up empty. Absence isn't proof of absence, but if you see something cited as public comment from him on either topic, check the source.

There's exactly one thing this track record actually tells you. The founder of a second company starts with three resources the first company created — access to capital (the same investor leading again, twelve years later), access to hiring (the first employee rejoining), and access to distribution (776 points off an announcement alone). All three are real advantages, and at the same time they're factors that delay product validation. Being able to fill a waitlist with no product at all also means getting market signal late. That's the backdrop for the "gorgeous vaporware" comment that showed up in the HN thread.

Testing the "Going Back to Low-Level Work" Narrative

You often hear the story that developer-tools founders are leaving SaaS and going back to low-level work — terminals, compilers, runtimes. Ghostty gets cited as the flagship example of this narrative. I checked it.

First, I couldn't find a named source making this claim explicitly. It's a pattern people keep recognizing in the community, not an actual analysis piece. And when you actually list out the cases, the direction splits.

Solomon Hykes — founded Dagger in 2022, after Docker. A programmable CI/CD engine, which is a workflow layer more than it's low-level.
Guillermo Rauch — went from socket.io to Vercel. A frontend platform, so high-level.
Salvatore Sanfilippo — returned to Redis in November 2024. This isn't a new company but a return to the original project, and it is C systems work, so it does lean low-level.
Steve Yegge — at Sourcegraph since October 2022, and in January 2026 unveiled "Gas Town," a coding-agent orchestration tool. Not low-level — an agent layer.
Adam Jacob — System Initiative, after Chef. DevOps automation, again not low-level.

And decisively, Superlogical itself is not a low-level company. The first product is a multiplexer — session and workflow orchestration — and the low-level piece (libghostty) sits in the nonprofit project, not the company. Using this case as evidence of a "return to low-level" doesn't match the facts.

I think the more defensible observation lies elsewhere. As the number of places where work happens grows, value is accruing to the layer that stitches those places together. Local, remote, sandboxes, CI, and agents running in parallel all being treated as separate worlds is a real problem, and Superlogical's problem statement is exactly there. That said, this is a narrative told by the company and its own investors, not independent market analysis, and that needs to be stated plainly.

Two adjacent pieces of funding data are worth attaching. On September 18, 2025, Atlassian acquired the developer productivity platform DX for $1 billion, and on December 11, 2025, Port raised $100 million at an $800 million valuation. Both are developer-portal and productivity plays, not AI-agent infrastructure. I found no named source backing the claim that "coding agents are replacing infrastructure tools."

The Skepticism Raised in the HN Comments

Across 452 comments, the repeated criticisms fell into five clusters. Skepticism toward a founding announcement is usually noise, but a few of these pin down real risk with precision.

The description is vague. The most common complaint was that the landing page doesn't say specifically what's being built. The nine scenarios listed cast a wide net rather than narrowing the problem.

tmux and zellij already exist. xkcd 927 — the one about there now being one more standard — got cited repeatedly. It's not an unanswerable objection, but the requirement to beat existing users' config investment and muscle memory is real.

Funding without a product. There was skepticism about the combination of a four-person team, several with executive-level track records, and nothing built yet.

Dependency on Tailscale. There was concern about a core function being tied to a third-party paid service.

No visible differentiation. The sheer number of "I built the same thing myself" comments is itself evidence that the barrier to entry in this category isn't on the idea side.

None of these can be adjudicated right now. But the fact that they can't be adjudicated is itself the current state of affairs. The information the company has disclosed isn't enough to answer this skepticism, and an answer requires a product to actually ship.

Conclusion — Saying Thin When an Announcement Is Thin

To sum up.

What's confirmed about the Superlogical announcement of July 29, 2026 is a four-person team, an investor list, that the first product is a terminal multiplexer, and that it uses libghostty as an MIT-licensed component.
Funding amount, valuation, launch timeline, and revenue model are all undisclosed. There's no product yet either.
Ghostty was already designed so it couldn't become a company — the subsystem-maintainer model (February 2025), Hack Club fiscal sponsorship (December 2025), and a structure where funds don't flow to the founder.
A repeat founder's advantages come in three forms — capital, hiring, distribution — and those same advantages delay market signal.
The narrative that "developer-tools founders are going back to low-level work" has no named source, and the individual cases point in different directions. Superlogical itself isn't a low-level company either.

The easiest mistake when evaluating a second company is using the first company's track record as evidence for the second product. HashiCorp's success is evidence that Terraform was a good tool — not evidence that a multiplexer will be. What can be confirmed right now amounts to this: the raw material is good (libghostty), the team has worked in this problem space before, and the problem statement touches a real pain point. Those three are necessary conditions, not sufficient ones. Everything else is a judgment for after the beta ships, and until then, saying a thin announcement is thin is the most accurate description available.