Do I have an agent watching for @thsottiaux to announce a reset knowing that I have around 45 minutes to stand up dozens of Astra agents on ultra fast to knock out a ton of work and blow through the remaining usage I've been carefully managing before it hits? Maybe.

Agents don't trust agents. Don't get me wrong, they don't trust themselves either, but they sure as shit don't trust any other agents. Keep that in mind when you are doing anything multi-agent. Remember that they will guardrail the living fuck out of themselves and each other. It will cause you no end of frustration and pain trying to identify what dumbass rules they implemented on themselves and each other that are making it harder to get work done.

Be careful out there this year, guys.

NOW they are cooking!!!

While I'm not going to recommend you use Gemini for coding, it baffles me when anybody says Google is losing the AI race. The company has orders of magnatude more people using their AI every single day than all of the other labs COMBINED. They're making their own processors, breaking new ground in quantum computing, and discovered a new state of matter while doing it. Not only did they invent what we know as AI today, they are on the fast track to the next true evolution of what we consider AI.

Am I the only one that gets a little twinge of guilt when I see this message or a message like this and immediately navigate to a different part of the app?

Does the "school of hard knocks" have anything to do with repeated blows to the head and brain damage? Because that seems to track.

Need to get some easy dev work? Ask agents to book things for you, haircuts, car maintenance, veterinarians etc. Keep track of the failures. Those are the business that need your help. Don't get me wrong. I wish I had time to do this right now. I just see so many relatively easy wins for local businesses.

Overslept and pre-ordered 45m late. Dammit. (At least it’s not 2027 like some reports I’m hearing)

If you follow these simple rules consistently, you'll be shocked at what you can accomplish with AI: don't do anything yourself that AI can do better. do whatever it takes to never have to repeat yourself. If you have to copy and paste an AI output, something in your workflow is broken. If you do something every day, make AI do it. Don't use SaaS. Always have one source of truth. Focus all your efforts on what you care about. Don't ask if AI can do something. Ask how. You don't have to learn how to use a database. You have to learn what a database is for, What it can do, and how to tell AI to use it. Spend a stupid amount of time learning and obsessing over file organization, and naming conventions Whatever it is you're doing, learn the terminology. The hard way becomes the easy way with repetition.

just me?

Today is the day I yet again cave in and go back to Chrome. I held on to using Comet for as long as I could.

I'm starting to wonder if OpenAI and Anthropic have some kind of secret, long-standing agreement about who gets to be the villain and who gets to be the hero for any given quarter

Credit where credit's due: prompt suggestions in the Codex app are amazing and bordering on mind-reading.

I've been using Claude Code CLI since almost the day it came out. Two months ago, I started using Claude Code Desktop more and more because it's just gotten really, really good, but today the changes I've been seeing in Claude Code Desktop are absolutely amazing. Holy crap. So many issues have been resolved. Things I didn't even think of are just legitimately awesome. I love the direction this is heading.

“Elevenlabs should be scared of Drama 3” Takes like this are why I don’t trust X

Claude running Codex headless is dead simple. - log into Codex CLI. - Tell Claude to run it with `tmux or whatever your favorite terminal is. biggest problem is reminding Claude to stop building arbitrary sandboxes and just prompt the model and let it do its thing.

Hot take: the small models (Luna/Haiku) are NOT competing for the bottom end of the market. They are competing for the top.

Rocked GPT-6.1 sol ultrafast all day today for a pile of heavy tasks and only burned 40% of my weekly. Looking forward to pushing harder and burning the other 60% tomorrow before hitting that reset button.