Rida Al Barazi

The most interesting part about this memo is that at its core, it's advocating for solid operating principles. This is how any engineering business should operate. The bottleneck elimination principle is just one example.

Great read. I’m lucky to be in the “empowered” camp.

I exhausted all my 4 (2 max and 2 pro) subscriptions this week in 6 days. I’m agent free for the next 24hrs! Too bad this didn’t line up with the weekend

A few techniques that helped me get better UI/UX results when working with coding agents: Having a design.md file or package. You can use Claude Design to set this up or have an independent Claude session just to build one from your existing codebase. For existing projects I asked Claude to Ultracode a workflow to scan and analyze the existing codebase and document the ui and ux patterns into a design.md file.  The front end development skill (from Claude) is a solid one too.  Asking the agent  to take screenshots and verify against them helps a lot (token extensive though) I also take screenshots myself and annotate them to provide feedback. There are some good tools and chrome extensions that help with this too. I have @benjitaylor’s Agentatuin set up on some projects. I’m yet to try @kitlangton’s flow: nitter.cf/kitlangton/status/2111…

Great read. I’ve been using around 70% of those patterns. Re: Cloud agents vs Local. My local set up has been pretty good across containers and different computers. The key is proper isolation. The “threads” callout is new to me. Maybe it’s an exclusive Amp feature.

If you run claude code and codex on more than one computer get one of them to analyze your usage patterns and errors they consistently see. You will find some low hanging fruit to fix and some patterns to extract into agent skills. Here's a cleaned up prompt I used after it collected the "field notes": Before implementing any recommendations, preserve the analysis and field notes you collected about my Claude Code and Codex usage patterns, recurring errors, and opportunities for reusable skills. Another agent on a second computer will receive this same prompt. The goal is to share findings, compare evidence, and develop improvements that work across computers and can also benefit other engineers. Please do the following: Save the findings in a shared repository. Start by checking whether the agent-skills repository is the right home and follow its existing conventions. Keep each computer’s notes separately identifiable so both agents can contribute without overwriting each other. Commit and push the notes so the other agent can access them. Separate observations from proposals. Document recurring problems, supporting examples, suggested fixes, and uncertainties. Distinguish broadly useful patterns from machine-specific issues and personal preferences. Recommend the appropriate scope for each improvement:Shared skills or tooling for multiple engineers and repositories. Repository-level instructions or setup for everyone working in a project. Machine-specific configuration. Personal preferences maintained through version-controlled dotfiles. Review existing setup mechanisms. Look for an existing repository setup skill or equivalent workflow before proposing something new. Avoid relying on untracked files in Codex or Claude home directories for behavior that should be portable or shared. Compare the two sets of notes when available. Identify common findings, differences, conflicting recommendations, and which conclusions have stronger evidence. If the other agent’s notes are not available yet, leave a clear handoff describing where yours are and how to contribute. For now, publish the findings and a proposed implementation plan. Do not apply the fixes or install new skills yet. Include the repository location, commit reference, and recommended home for each proposed change.

Finally!!! If this works as good as advertised I’d get rid of my macgyvered solution right away

Disaster recovery plans and credential rotation drills shouldn’t be plans and drills anymore; they should be part of your CI and CD. Practiced at every release. If we can guild ease now then we should be able to recover as fast as we build. This is the new normal.

We don’t need GitHub anymore! GitHub was built for humans to look at each others code and protect releases. We (humans) no longer “write” the code. We no longer look at the code, we have automations to verify it better than any humans would’ve. We can assess its risk and then ship. We don’t need GitHub anymore. We need a source control with hooks and we don’t even need a ui for it.

I'm not alone! I guess this is why most of my side projects turn into infra and dev ops projects within a few days of hacking.

I asked @ChatGPT to make a math practice worksheet for my daughter (grade 4). It first created a crappy html page that didn’t work. Then it dynamically rendered the UI inline with a personalized game and it worked perfectly! My daughter said: “beat that IXL!” (That’s the ancient software they assign homework on).

Git was built for a different time. Not the time of agents that are always online!

It took less than a day and now my wife can edit (code and deploy) the dashboard herself using her ChatGPT account! I signed her to GitHub and added her to the repo which has CI/CD and auto updates the kiosk app on the frame. Now she’s talking to ChatGPT and redesigning the whole thing to match our family’s aesthetics 😂

Family dashboard live and complete. Plugged the photo frame into my laptop last night and used Codex remotely to root it, build the site, host it on my hetzner box, add some jokes and activities fun for my daughters, secure it… geeked out on the whole thing! Thanks again @wesbos for the idea!

Can’t stop thinking about all the possibilities to repurpose all the old hardware I have. Old kindles, iPads, a couple of MacBooks and a dell computer. Between Omarchy and rooting them as repurposed screens. Just unleashing Opus 5.5 with Ultracode and experimenting!