Joshua Warren

A useful AI workflow starts with a concrete recurring decision, not a vague innovation project. Daily competitor checks are a strong early win because they are structured, repeatable, and immediately actionable. That’s the takeaway from this Commerce Today clip. #Ecommerce #

The highest-ROI AI move usually is not the most glamorous one. Start with repeatable internal workflows where AI saves time, improves consistency, and compounds every day. That’s the takeaway from this Commerce Today clip. #Ecommerce #AgenticAI #AIShopping #DigitalCommerce

Read the paper at github.com/joshuaswarren/… + update to the latest version of Engram github.com/joshuaswarren/… to experience this new technique today. It will help your agents meet your expectations the first time and when your agent does make a mistake, it will learn from it quickly

From this research, I've created a new approach to agentic memory - Prediction-Error-Driven Calibration (PEDC) for model-user alignment in agentic memory systems.

My goal is to make it easier for users and agents to work together, specifically for one user and one agent to learn how to best work together. I've been researching the frontier of agentic memory to push this forward.

My Engram memory plugin for @OpenAI Codex and @OpenClaw has focused on bringing existing SOTA memory techniques to these platforms. Until today...

Customers do not want novelty for novelty’s sake. They want fewer clicks, less confusion, and faster decisions. The best AI use cases remove friction instead of adding another flashy feature. That’s the takeaway from this Commerce Today clip. #Ecommerce #AgenticAI #AIShoppin

@thsottiaux I'd love to try them out...on March 18th at 11:48AM, it looks like. Unless there's some reason that usage limits need to be reset before then...after all, it must be a bug that I used my ChatGPT Pro limits that fast, right? 😁

Hello subagents in codex. Have seen some awesome new and creative workflows emerge from these developers.openai.com/codex/subagent…

AI-assisted product research moved from edge behavior to mainstream far faster than most merchants expected. Merchants need to rethink discovery, merchandising, and conversion before customer behavior outruns the stack. That’s the takeaway from this Commerce Today clip. #Eco

@vincent_koc @nvidia @steipete Excited about this, but also a little nervous about how big of a rewrite is ahead of me for my memory/context plugin. It'll be worth it in the end, though - can't wait to see this update!

Finished shipping at 2am at @nvidia for GTC alongside @steipete, to sleep and repeat another crazy week in AI. But most importantly a huge upgrade to the claw coming soon… 📦 First-class plugin architecture 🔒 New sandbox and telemetry for saftey 🧑‍💻 Codex app server support 🤌 Native Claude and Codex plugin support 📟 Multipass for E2E channel testing 🔭 Updated Opik Openclaw Integration Stay tunned!

@youphilthis @_dmca @alexkehr @mlg27_ Codex and Xcode only. Well, and some Xcode MCPs and skills.

@JoshuaSWarren @_dmca @alexkehr @mlg27_ What stack are you using to build iOS apps? Codex and Xcode only?

congrats to @JoshuaSWarren 🎉 check out his iphone app in the reply. kudos as well to @alexkehr and @mlg27_: both very solid runners up 🦾

I didn’t even realize they added that UI - I really need to update! It’s been super useful since I stumbled across it a few weeks ago - keeps skills in sync across all my agents across two computers. So nice not having to think “is that skill installed in codex or just Claude?”

@cathrynlavery I've found skillshare to be super useful for this - github.com/runkids/skills…

💻Vibe coding pro tip Your AI skills shouldn’t live in 3 different places that create redundancy. → create a .agents GitHub repo → put ALL your agent rules, skills, personas, scripts, prompts there → symlink into your Claude Code, Codex, Cursor, etc. When you change one file, every tool stays in sync. Also your @openclaw can use the same skills too 🦞

@_dmca It’s a toss up between Done and Openclaw Engram - the memory system I’ve built for Openclaw. ChatGPT Deep Research + Codex have allowed me to go deep on understanding advancements in agentic memory and then bring them to Openclaw! github.com/joshuaswarren/…

Happy Birthday! Codex allowed me to build and launch my first iPhone app after being in the Apple developer program for over 15 years but never finishing an app! Ironically, it’s an app to help those of us with ADHD get things done. apps.apple.com/us/app/done-ta… Finishing up another app for the neurodivergent tomorrow, too!

it’s my birthday tomorrow. let’s celebrate. reply with the coolest thing you built with codex. winner gets 3 months of pro.

I built this with Engram -a local-first memory plugin for OpenClaw. Plain markdown files. YAML frontmatter. No hosted service, no mystery database. Your memory stays on your machine. Free, open source: github.com/joshuaswarren/…

The uncanny moment isn't "wow it remembered my name." It's when your AI reminds you about a blocked Stripe integration test while you're asking it about gardening. Context that shouldn't cross-pollinate — but does, and it's useful.

My @openclaw agent knows I grow basil and tomatoes. It also knows I'm tracking 37 concurrent projects and that my CRM sync cron has been failing to post to Discord for 3 days. It knows both of these things in the same session. That's the weird part.