I think @AnthropicAI is having problems. I just got this message: "here's an issue with the selected model (claude-fable-5). It may not exist or you may not have access to it. Run /model to pick a different model." Buuuuuuuut, been using it all day, so...? 🤷🏽‍♂️

I just a new issue I don't seem to have recalled hitting with Codex. It refused to give me data out of my own database. I asked to inspect how a password was stored and retrieve the value. Something anyone with admin privs can do. Which I have. Not only did it refused, it marked the conversation for their trust and safety team. Nothing will cause me to absolutely 100% refuse to use an LLM than for it to refuse work on my own projects. My follow up prompt was that I had admin access and to use the key that I have provided in a separate file and it still refused. This puts Codex in the coding penalty box for me. I don't care what "best practices" are for data security. If a value is in the db and the user's db account can see the data, show me the data. The model should not refuse to take an action when credentials are presented. This breaks a core tenet of using products: Customer Trust. This eroded a ton of trust. Worse, I don't want to ask the same of Claude because I also know that it will likely do the same. Why are we nerfing these models in arbitrary ways if they are meant to help us work faster?

of all the really super duper annoying bugs in Claude Code: I really hope you don't accidentally hit ctrl c instead of ctrl v when pasting something into a large block of prompt text. Congrats - it's all gone. y tho?

I can say this with 100% certainty. There is no chance I get through the CASA 2 security audit for my new application without Codex and Claude Code. No. Chance. This process has taken far longer than I imagined just to have Gmail API access. I should make a whole video about how the agentic harnesses allowed me to understand the scope of the coming audit, use open source tools to prepare for and mitigate any issues ahead of the audit, and pass on my first try. Anyone who says agentic coding is a sham is a moron. The entire application is 109% agentic built. How we build has changed. The speed at which we build has changed. That possibilities of what a motivated individual can build has changed. Forever.

I've never experienced the rage @DNVR_Rudo has for missed faceoffs during playoffs. We missed a goal too. I have rage.

Well, that's a first for me. Wrote this to Claude: " This has been going for a **VERY** long time - please write up what you are trying to accomplish that is not yet done in a prompt I can hand to another agent. You are stuck." Codex to the rescue?

agentic based playwright testing is 100% saving my sanity as these apps get bigger and bigger.

Claude had an outage earlier this evening. Working with Opus 4.7 right now feels like a lobotomized intern. I have to stop working on everything right now that Claude is touching. The mistakes that are being introduced belie credulity. One example: something is broken so the fix was **TO DELETE ALL CACHING** for my app. Is this a fix? Technically, yes. Is it the right answer? Not even remotely close.

Now that I am paying fort eh $100 plan on Claude and Codex, if I think one of them is about to really do something wrong, I use this prompt. "I don't agree with you. Write up what you think is wrong so I can ask a different agent." This has been a huge stress reducer for me.

To whom it may concern at @NHL. You have three games today. Exactly three games. Exactly three playoff games. How is it that even today has your most valuable games on the schedule that you can't make it so none of them run over each other, but in the regular season we'll get a 27 minute delay for puck drop because of some random game happening ?

This is the first time I have seen programmatic SEO work for anything I have built. I have no idea if this is an anomaly or if this continues, but at least I get to smile for a few hours.

I wrote a skill to closely mirror (though enhances) the /insights tool in Claude Code, but added in ability to review all Codex work and write a single report. It is built in a way that can begin execution from either, but run the other headless to continue working on the report output. If anyone expresses interest, I will publish to a git repo. Here's what it does: Each agent reads all my sessions from the last 30 days, writes an HTML report, then auto-invokes the other agent's CLI (codex exec / claude -p) to append a peer-reviewed Second Opinion. Both agents coach me on how I'm managing them for things like delegation quality, checkpoint discipline, escalation thresholds, and things I can do to improve my agentic coding. As a funny example, the skill told me 43% of my Claude spend this month went to sessions where I kept giving Claude "one more chance" instead of handing off to Codex. Lolz. 🤷🏽‍♂️🤦🏽‍♂️ The unlock isn't the stats. It's that the skill is framed as coaching for managing agents, not just usage reports. How do I delegate? When do I escalate? Where are my skill-portfolio gaps? It turns out that my biggest habit problem is tolerating Claude loops far too long before reaching for Codex. I have the escalation tool, but I don't reach for it often or fast enough.

Not me at all being a menace

Anyone in my network have pointers or process helpers for passing CASA 2? I finally passed the other items for Google API restricted and sensitive scopes and must now do CASA 2. I have never done one of these before.

Good jokes will never ever go out of style

This is what my life has become... "wrote: google_oauth_compliance_corrective_action_workplan.md" So. Much. Fun.

Imagine being a company that wants to organize the world's information, and has perhaps the largest amount of compute available to any company. Now imagine that it takes them 5 days to turn around each step of the API authorization process for every "we need this" step and they only give one thing at a time. I am not almost into month 2 of trying to get API access for something that took me 4 weeks to build. This is so crazy. Meanwhile I cannot make money because they are the gatekeeper. I have to keep turning away customers. Bananas.

Over the last 48 hours I have learned that a) getting your sitemap write for programmtic pages is hard, b) google doesn't really help much with Search Console, and c) @AnthropicAI Claude Code will happily waste time and energy on what it calls "guesses" any time you point out that what it asked you to do did not work.

As with everything in life: timing is everything