The Stack: what we’d actually install on a new machine in 2026

A blank machine is the only honest benchmark. Not what you recommend at parties, not what you defend in a comment thread — what you actually type into the installer in the first twenty minutes, because you have work due.

That’s the premise of The Stack, a recurring column about working tools. Ground rules, stated once and binding forever: every pick links to the tool and names a credible alternative; every verdict is falsifiable. We don’t run a test lab. Our verdicts synthesize the published record, named and linked, plus our own daily use, and when a claim is someone else’s reporting we say whose. No affiliate links, ever. If a pick ages badly, the correction goes in this column.

Here’s the 2026 install list.

The browser: Firefox

This one got easier, and Google did the work. Starting with warnings in late 2024 and rolling through 2025, Chrome progressively disabled uBlock Origin and every other Manifest V2 extension as part of the Manifest V3 transition, a rollout BleepingComputer tracked wave by wave. Firefox kept supporting the extension architecture that full-strength content blocking needs, a stance TechRepublic contrasted with Google’s at the time. A browser is the room you spend eight hours a day in. We’ll take the one where the ad blocker still works as designed: Firefox, plus uBlock Origin, before anything else touches the disk.

The credible alternative is Brave, which builds blocking into the browser itself and so sidesteps the extension question, at the cost of Chromium underneath and a crypto sidecar you’ll want to switch off.

What we’d skip despite the hype: the AI browsers. We watched Arc, arguably the most loved browser of the decade, get feature-frozen while its maker chased an AI pivot; Atlassian then bought The Browser Company for $610 million in cash in September 2025, as TechCrunch reported, with development effort long since shifted to Dia. Every user who built their tab muscle memory around Arc ate that decision. We’re not doing it again for a browser whose main feature is summarizing pages we were about to read anyway.

The editor and the terminal

The editor is Zed. It’s built in Rust by Nathan Sobo and other veterans of Atom, open-sourced in 2024 under a mix of GPL, AGPL, and Apache licenses. Since Windows support landed in October 2025, it’s finally recommendable across all three platforms. It opens in a blink. The AI features are optional rather than load-bearing, and you can read the source. One caveat we’ll keep watching: Zed raised a $32 million Series B led by Sequoia in August 2025, and venture-funded developer tools have a documented habit of growing login screens. The licenses are the insurance policy.

The alternative is the obvious one: VS Code, still the default for a reason, with an extension catalog nothing else matches. If your workflow lives on three obscure extensions, stay put. That’s not a defeat; that’s the trade.

The terminal is Ghostty, Mitchell Hashimoto’s post-HashiCorp project, which hit 1.0 on December 26, 2024. It’s fast and platform-native, with defaults you won’t need to fight. Hashimoto wrote about why he built it when terminals already existed:

I saw tradeoffs that I didn’t like. I saw features that I wanted. I saw performance that I could improve. I saw stagnation.

He also wrote, in the same post, that “Ghostty will always first and foremost be a passion project for me.” In a category where the best-funded competitor, Warp, shipped for years with a mandatory account (a requirement it only lifted in late 2024), a passion project with no growth team is the feature. Alternative: WezTerm, if you want your terminal configured in Lua and endlessly scriptable.

The AI assistant: rent, don’t marry

Here is where a tools column has to be honest about its shelf life. The working answer in mid-2026 is a terminal coding agent, and there are two serious ones: Anthropic’s Claude Code and OpenAI’s Codex. The published head-to-heads, Builder.io’s and DataCamp’s among the more careful, reach split, heavily hedged verdicts, and the underlying models leapfrog each other every few months.

So our pick is a posture, not a product: install whichever agent rides on the model subscription you already pay for, keep your prompts and project notes in plain files you own, and assume you’ll switch within a year. Anyone who tells you the 2026 assistant question has a permanent answer is selling the answer.

Notes and passwords: boring on purpose

Notes go in Obsidian. The pitch hasn’t changed in five years: plain Markdown files on your own disk, readable by any editor that will ever exist. What changed is the price. In February 2025 Obsidian made its commercial license optional, so it’s now free for work too. “No account required, no ads, no tracking, no strings attached,” as the announcement put it. The alternative is Notion if your notes are really a team database wearing a trenchcoat. Ours aren’t.

Passwords go in Bitwarden: open source and independently audited, with a free tier that covers most people and a premium tier at ten dollars a year. The alternative, 1Password, is genuinely more polished, at roughly $36 a year and closed source; comparison testing like Cybernews’s consistently gives it the ease-of-use nod. We’d rather have the auditable one, and we’d rather you use either than neither.

Tally it up. Firefox, Zed, Ghostty, Obsidian, Bitwarden: free, and every one of them either open source or storing your work in files that don’t need the company to survive. The AI agent is a rental by design.

Total defensible recurring cost of the 2026 stack, beyond the model subscription you were paying for anyway: ten dollars a year.