Flox

A lot of threads right now are arguing over which AI coding agent has the best reasoning capabilities. But if you watch them execute, they almost always hit the exact same wall... So what do Cursor, Claude Code, OpenClaw (and it's variations), Ollama, Antigravity, Copilot, Codex, CodeRabbit CLI, Qwen-Code, and the Flox Model Context Protocol Server have in common? You can now spin them up turnkey. We just launched agentic-development-with-flox out of closed preview. If you want to look at the actual engineering reality behind this... It’s rarely the generated code that fails. It’s the environment, and the complexity of setting it up. The agent hits a missing C++ compiler or a Python version mismatch, and immediately enters a token-burning death loop of apt-get and pip install guesses. It hallucinates infrastructure fixes until your API budget is cooked and your host machine is a polluted mess. I feel like judging an agent's capability inside a mutable, unpredictable environment is completely missing the point. We are asking them to solve configuration drift, not write software. I think the only way to unlock actual agentic development is through mathematical determinism. (#NixOSNixOS) We just published the guide that goes deep into how we made this turnkey. Link below. Back to the wasteland, more soon! 🦞 🏜️

We spoke with Farzan Karimi (Deputy CISO @ Moderna, SunsetCon) about what’s changing in #security right now: - Automated attack simulation is becoming standard - AI agents are taking over low-level pentesting + recon But the real shift? Agentic offense + agentic defense, working together. - continuous attack simulation - real-time detection tuning - automated ticketing + remediation loops This is where security engineering is heading.

#AgenticDevelopment Dependency resolution, version drift, conflicting toolchains, “works on my machine” → all of that burns tokens and breaks workflows. This is where Flox fits. 🟣 Cross-language, cross-toolchain package manager 🟣 Works across OS + CPU/GPU architectures 🟣 Deterministic, declarative environments (powered by Nix) 🟣 No containers, no VMs, runs directly on your system Agents (Claude Code, Codex, Gemini CLI, Copilot CLI, etc.) can: 🟣 Discover and install exact dependency versions via MCP 🟣 Skip dependency resolution entirely (it’s already solved) 🟣 Run conflicting versions side-by-side (e.g. OpenSSL 1.1.1 + 3.0.x in the same runtime) The repo in this article is a practical reference: 50+ prebuilt, turnkey environments for: - coding agents (Claude Code, Codex, OpenCode, etc.) - multi-agent orchestration - local inference (Ollama, vLLM, Triton) - full dev stacks (DBs, schedulers, services) Read the full guide + grab the repo! Linked in the comments 👇

Super fun bringing together such an exceptional group of engineering leaders in NYC last week. Grateful to partner with Jai Malhotra and Mustafa Neemuchwala to make this happen, and looking forward to many more. If you're ready to defend some hot takes and based in the city, join us next time. OpenAI New Enterprise Associates (NEA) Flox The First Commit

In this PlanetNix short, Tom Bereknyei, a long-time contributor and steering committee member, breaks down where #Nix really shines and where it doesn’t. If you’re working across: 🟣 multiple languages 🟣 different systems or architectures 🟣 constantly changing environments Nix lets you glue everything together without losing control. From consulting across clients to building CI and automation systems, this is where Nix becomes indispensable.

If you've ever said "it works on my machine," this one's for you! Flox is a package and environment manager that makes your dev, CI, and production environments actually the same. Declarative TOML, powered by Nix under the hood, no containers required. They're this week's newsletter sponsor! What that means practically: you define your environment once, and it resolves to the exact same dependencies on every machine, every time. Linux, macOS, x86-64, ARM — it doesn't matter. Your environment travels with you. There's also a really interesting angle here for AI-assisted development: agents need reproducible environments too. Because Flox environments are just TOML, agents can inspect and modify them directly. Agentic workflows get the same 190,000+ packages humans do, with the same conflict resolution primitives. If you want to go deeper, the team wrote a great piece on using Flox as a turnkey toolkit for agentic development: https://fandf.co/3NRXoGy

We recently went live on LinkedIn with William, Michael, ✨ Jacquie Capur, and Kelsey Hightower, and got into how #AI is changing the way we build. You can spin up a prototype in no time, try a bunch of ideas, throw them away, and start again the next day. That used to take real effort and now it’s kind of the default. But just because we can build more, faster… doesn’t mean we should stop thinking about how we’re building. So the real shift isn’t just “build faster with AI.” It’s zooming out and asking: how does all of this fit together? → how AI plugs into your workflows → how you keep things consistent as things change constantly → how you avoid chaos as the number of moving parts explodes That’s where having a solid, reproducible foundation starts to make a big difference. Check out the full webinar here 👉 https://buff.ly/GWafnp6

Open source only works because people can share work, build on it, and trust that what’s being handed off is something they can actually reproduce. That idea matters even more now, when AI is changing how software gets written, reviewed, and contributed. 🟣 What happens to open source when maintainers are flooded with AI-generated PRs? 🟣 What does reproducibility mean when the output changes from run to run? 🟣 And how do we keep collaboration healthy when the barriers to contribution are dropping, but the burden on maintainers keeps going up? This session was a really thoughtful look at where #reproducibility fits into all of that, not just as a technical property, but as part of how communities work. Great conversation from James Bayer and Stormy Peters at PlanetNix 2026 ↓ Watch it here 👉 https://buff.ly/9q5p19j

AI is accelerating your software delivery but it’s also increasing risk, complexity, and accountability. Instead of chasing drift, debugging inconsistencies, or relying on post-hoc fixes, you get deterministic environments that behave the same from dev → CI → production—every time. That means: 🟣 Stable, available systems, even as change velocity increases 🟣 Full visibility from training → evaluation → production 🟣 Audit-ready provenance, SBOMs, and faster CVE response 🟣 Faster time to value with fewer regressions 🟣 Lower costs from fewer rebuilds, reruns, and rollbacks Velocity without control isn’t a strategy. See how leading teams are building with both → https://buff.ly/jjYPkDU

What makes #Nix click for a lot of teams? At PlanetNix, Jacek Galowicz pointed out that a lot of teams are focused on moving fast, scaling things, and figuring things out as they go, rather than over-planning everything upfront. And Nix actually supports that way of working in a very practical sense, because it lets you test complex systems in a way that stays simple, fast, and reliable. Shoutout to Jacek and Jacquie for a great conversation! #PlanetNix