5/ If you're tired of struggling with email testing - or lack thereof - give MailMock a try. It's designed to seamlessly fit into your workflow and minimize the usual email mocking overhead: mailmock.dev

@KingBootoshi Sounds like 5.5 medium or higher. It’s nice for planning but for the implementation it has a tendency to go off on rabbit trails and attempt to handle more than it should.

Ok. Codex cannot orchestrate itself. It cannot prompt itself. It cannot control itself. It is an ouroboros. A powerful serpent of pure might that eats its own tail if it is given complete freedom. It fucked up so bad today - to the point where I am considering going back to using Claude Code as the ORCHESTRATOR of Codex Agents (very important key distinction there) I am happy I used Codex exclusively for the last couple weeks, I understand its limitations, strengths, and weaknesses even more, and it's exact purpose where it fits in my workflow: The autistic coder that follows directions and generates good code like an absolute champ Claude Code, the model that can understand human intent like no other, and prompt/direct other agents like a TRUE manager, is going to take back its place as my direct co-pilot. Now, it's possible today's fuck up is entirely my fault (it partially is, I'm not sure if I should take complete responsibility, but I'll leave you to judge that) Here's what happened: - I am ripping out Supabase out of my app, in replacement for Postgres directly - First, I had /goal implement the Postgres base. Works great, made it's own Postgres system - did NOT replace Supabase directly yet. I wanted to get the base foundation working - We replaced core primitives with this Postgres + new auth flow (since we used Supabase for auth too) and dogfooded it to make sure we can get through onboarding using this new flow, minding other features - We even replaced Supabase storage with MinIO in preparation to host our prod on AWS - The CORE functionality, and loop of the product, was successfully (and quite sexily) implemented. After 6 hours this was complete - Core functionality (as instructed) was completely replaced with this new infra. Tested it - awesome Here's where the fuck up happened: Then, the next /goal was simple: "Completely remove Supabase from our app while maintaining base functionality" After 3 hours of work, I glance at the logs and I see Codex narrate: "Now that we removed Sentry" What? Why the hell did Sentry get removed? That was NO where in the scope? That is our performance monitor. I DEFINITELY did not ask to replace Sentry. What happened??? I looked at the traces and the logs - the Sentry npm package we had installed, had A SUPABASE DEPENDENCY inside THE NODE_MODULE ITSELF CODEX SAW THAT SENTRY'S PACKAGE HAD A SUPABASE REFERENCE, AND COMPLETELY REMOVED THE ENTIRE DAMN THING DAWG NO I STOP IT IMMEDIATELY Then I ask, with fear in my prompt: "What else... did you remove?" > it removed every single feature that didn't get ported into Postgres besides the core product loop > bruh Since we didn't port everything to Postgres completely (since I wanted to ensure the base foundation was designed properly) It followed instructions to the T. It removed Supabase completely - and every single feature that had some involvement with Supabase that's how AUTISTIC and GOAL DRIVEN this ABSOLUTE ERDOS SOLVING GENIUS OF A MODEL IS and this is precisely the lesson in HOW careful you have to be with prompting: there can be no room for vagueness in your goals, in your prompt, and in your direction because if there is even the slightest Achilles heel in your prompt assume that it will be slashed the funniest part, is this would've been flawless if the goal was more focused: If we did this workflow instead: 1. Completely replace ALL app functionality with our new Postgres core THEN 2. Completely remove only *WRITTEN* Supabase functionality from our backend It would've been flawless. Because I'm ngl it did a damn good clean job of ripping everything out 😭😭😭 anyways guys that's why we commit and save checkpoints! codex is a fucking glass cannon so just be careful LOL

@nizzyabi Many teams are asking this same question

why can’t i just use an agent to design my app/site in code? i don’t understand the added layer to designing then transferring that into code.

@burkeholland What’s the basis for some of these new model multipliers? 14x is very surprising for a Gemini flash model. Is there a formula or something, or is this a stopgap until usage based billing takes over?

well hello little fella

@samhenrigold lol

be careful out there, i just got mugged at the intersection of liberal arts and technology. i wouldn’t go out there at night

@TT11986 Most Valuable Phreethrow-shooter

son are the four pics... at the freethrow line?💀😭

@slicknet That is the best argument, but in the grand scheme of things I don’t feel like is a good enough reason to move off.

Best argument I’ve seen to move off of Tailwind classes in HTML.

@Y40IFRQTTING @thamosdeaf a) What would this fix? b) What would happen to those houses?

Fix everything instantly button: Make it illegal to own a house you don't live in

@maddymontaquila @code @aspiredotdev I love that the agent can see and use the built in browser. It’s super nice for having the agent manually test its work in the frontend.

i used to agree wholeheartedly but tbh the embedded browser in @code is nice for specifically the standalone @aspiredotdev dashboard to see Copilot Chat otel lol

@danveloper 2 bit on 128gb feels like it wouldn’t be the greatest combo

In case you aren't following - the Redis guy is smashing DeepSeek V4 into a local runtime to run on Apple Silicon. Big models on small computers is the way!

@NateSilver538 Vibecoding is everything you mentioned above, except for "reviewing code". The phrase "vibecoding" insinuates that you are more concerned about the vibes/output than the source code.

Q for a forthcoming story. Let's say a knowledgeable programmer writes a complex script with help from AI. He orchestrates the model and carefully monitors its progress, including reviewing code. But he isn't writing in code himself; the AI translates from his natural language.

@NateSilver538 They decided to start playing their bench in the playoffs.

I mean they've probably had 98th percentile shooting luck but sure seems like something clicked and this is a different sort of Knicks team with a higher ceiling than most (including me) envisioned.

@claudeai @SpaceX Interesting

We’ve agreed to a partnership with @SpaceX that will substantially increase our compute capacity. This, along with our other recent compute deals, means that we’ve been able to increase our usage limits for Claude Code and the Claude API.

@dosco As much as I despise writing YAML, I can see the benefits for use with AI for both input and output.

been saying this for years

@theo This has been a lot of fun to take advantage of. What would you move to once they switch to usage-based billing?

I sent a single message on Copilot and it did over 60m tokens. It's still going. $30 of inference so far. In their current billing model, you get 1,500 messages, regardless of how expensive each is. I'm pretty sure I can do $45,000 of messaging on this plan

@kenwheeler Imagine how many agents you can kick off in this bad boy

wife got me 2026’s hottest programming accessory for my birthday

@BarstoolHubbs @EvanMiya what would you call a 71-16 run? 😂

After this Hawks three I got that “oh boy here we go feeling” and then the Knicks answered with a prompt 71-16 run What a night

@MFrancis107 Let me know when you see them 👀

Are all of the OpenClaw Mac Mini's on ebay yet?

@pierceboggan @code Is there a timeline for the Agents app moving to the stable release?

OpenAI models are now 12% faster in @code with our recent move to WebSocket mode on the Responses API! #_websockets-supported-for-openai-models" target="_blank" rel="nofollow noopener">code.visualstudio.com/updates/v1_118…

@morganlinton I went on a similar journey, and so much more of my feed is making sense! All the acronyms make things super nice for tweeting but super hard to decipher if you don’t know what half of them mean 😂

Intelligence alone isn’t all it’s cracked up to be. With both humans and AI.