Mark Fox

@ross_cefalu @FactoryAI Does it support missions now?

Droid is now easier, faster, and more enjoyable to use in the Factory Desktop App. Excited to share the first thing I've been working on since joining @FactoryAI

@OscarFalmer metaglass.dev I'm updating any I find here to build out a central resource, anyone building can tag me and I'll keep adding

🕶️ Check out the Display Web Apps the community has built over the weekend! On Thursday, we released the ability to create Web Apps for Meta Ray-Ban Display and we are delighted to see the enthusiasm in only a couple days.

@krzyzanowskim This is underrated, and won't solve the issue but it's another tool in the toolbox to help github.com/imbue-ai/vet it parses session + diff + initial goal to determine drift.

"skill issue" - a whole day of writing PRD/specification - half day grilling the PRD/specification - 16h /goal implementing PRD it doesn't work. IT DOESN'T WORK. it doesn't work at all, but also it doesn't work as specified. why broken? "So implementation drift" I'm done with this shit!

Ok cool, thank you. If I wanted to update the main work folder itself though I need to in terminal? Example is.. working on multiple worktrees, merge, then I want to run main workspace to check everything together. Right now it doesn't update main only if I create another workspace?

@M_J_FOX @superset_sh Hey mark, We auto fetch now so your workspaces are always fresh from origin main without needing to pull (in v2) Plan is to have a smart PR/git button that’s much less low level and manual than what we have in v1

Back by even more popular demand, the run button. Also a little sexier. Now in the latest version of @superset_sh

Been running a @FactoryAI mission for 7.5 hours straight. 100M+ tokens and counting only halfway done. Tomorrow I'll share what came out the other side 🚀

@peterom Hi Peter, I have used both desloppify and megaplan, I truly appreciate what you're doing, is there a way to sponsor, provide support to your work? Or do you freelance etc? I'm looking for someone to assist me in agent orchestration projects. Thanks again

I'm too poor to run megaplan evals on proper models so I'm trying get impressive results from the cheapest possible open models I was able to beat Opus w/ a combo of GLM + Qwen on NextJS's eval (76% vs. 71%) but now I'm shooting for SWE-Bench - 16 problems in it's at 100%!

@peterom latest version is 🔥 it has been running for nearly 18 hours, on 91st wave now, real material changes across codebase.

@M_J_FOX Thank you sir, new version coming today

Introducing Desloppify v.0.8. Thanks to many workflow improvements + new agent planning tools, it can now run for days on end - autonomously finding, understanding, & fixing large and small code quality problems. There's no reason your slop code can't be beautiful!

@ojaskandy MOON

We just hacked macOS to build Moonshot, the first AI agent with its own computer on your Mac! Moonshot works as a separate macOS user alongside you. 2 users, 1 mac. Join the waitlist for early access at moonshot.computer comment "MOON" for priority access

@LLMJunky Chain-of-thought reasoning is comparatively vague and subjective. It’s less willing to push a task all the way to completion and seems quite resistant to delegating work to sub-agents.

GPT 5.4 has been out for a few days now. So, what's the verdict? How are the vibe benchmarks going? I've been traveling and havent had the chance to play with it. Just a wee bit jealous. Honest opinions? 👇

It’s damn scary if you’re not used to this. But the UAE has world class protection systems and I trust the government will look after everyone here. Our instinct is to run or do something, but the best thing right now is to stay calm, stay where we are, and carry on. The interesting psychological thing here is that humans are wired to act under threat. Evolution rewarded movement. That mismatch between instinct and reality is exactly why moments like this feel so unsettling.

update on dubai/uae why is everyone lying to you? and what’s next? #IranIsraelWar

GPT-5.4 in Codex now has experimental 1M token context. To try it: model_context_window = 1000000 model_auto_compact_token_limit = …

Warp Engine Alpha is an experimental extension for the pi @badlogicgames AI coding agent that fundamentally changes how the agent handles "memory" and context. Instead of the traditional approach where an AI assistant accumulates a massive, ever-growing chat transcript (which inevitably leads to context bloat, staleness, and hallucinations), Warp Engine replaces conversation history with an event-sourced architecture. mjfox.ai/docs/warp-engi…

@pvncher @RepoPrompt I jumped in early and it’s been great to the app evolve. If you had a moment you’re someone who I’d appreciate any thoughts you have on some of the experimental context systems I’m building out mjfox.ai/docs/warp-engi… very much alpha stage, agent Xena in my docs is worth a read

When I started @RepoPrompt it was really just about making working with web chats easier Along the way, I discovered that building good prompts, and managing context is essential to getting the most out of reasoning models. That still holds true with agents!

Adding context signals to "Noktua" a Mac desktop agent extended from @LangChain #openwork 👉 mjfox.ai/docs/noktua will make git public next week..

Simple pi agent prompt loop that works with Linear github.com/nof0xgiven/ass…

@OsaurusAI @OfficialLoganK for sure! I will send over a loom later today, thanks for the quick reply

@M_J_FOX @OfficialLoganK Thanks for trying it out. Could you tell me some pain points? Your feedback would be greatly appreciated

We made a skill for the Gemini API! github.com/google-gemini/…

@OsaurusAI @OfficialLoganK I tried it and really wanted to like it, the concept is on point. I’m excited for where you’ll get it to in a couple of months. If there is anything I can do to sponsor, help, LMK I would love to see it evolve and use it daily.

@OfficialLoganK I needed this while building this x.com/osaurusai/stat…

@itzlassse downloading now 👏

A basic agent mode btw, gonna improve this!! This is just the start