I hope there’s some step-change model on the horizon from @AnthropicAI because Opus 4.8 has been lazy af over the last 36 hours, just handing work back to me in an alarming quantity.
@mattpocockuk Def recommend @orca_build; worktrees, split panes, built-in orchestration, browser and computer use. Become my every day, all day driver.
@radu_cojocaru@orca_build@kilocode All of the above. I use orca on a headless mac mini sitting on my desk for personal stuff and then orca also has remote server functionality that sits on a VPS (https://t.co/9lykttWQj4). Promise this is not an ad for orca, just a big fan of what it's done for my agentic setup.
Feeling pretty bulletproof with my engineering flow right now: @orca_build ADE (*chef's kiss* on the orchestration); Fable 5 (med) as planner; 5.6 Sol (xhigh) for execution, Grok 4.5 (high) code review. THEN @kilocode PR review on github using Qwen 3.7+. Am I missing anything?
I hope there’s some step-change model on the horizon from @AnthropicAI because Opus 4.8 has been lazy af over the last 36 hours, just handing work back to me in an alarming quantity.
My assistant going down because a power outage left my Mac mini off is a boring infrastructure problem, not an #AI one. Def not something that gets fixed with a better prompt or a different model; it's autorestart settings, a UPS, health checks, and fewer heroic assumptions. The craft is often exactly this mundane; less "make the model smarter" and more "make the system survive a random Tuesday."
The thing I’m learning about the AI work that actually changes an agency is that it doesn't start with the model choice, or even the inference layer as a whole. It starts with a painfully specific sentence: "where does this work go when it is done?" If nobody can answer that, the demo is theater. Useful, maybe. Operationally irrelevant, definitely.
Today, as every day, we remember Diogo Jota and André Silva, who tragically passed away one year ago.
Through immeasurable loss and incalculable pain, the impact they made and the legacies they left behind - not only within the footballing world, but in the hearts and minds of so many around the world - has shone through over the last 12 months.
All of our love, support, thoughts and prayers continue to be with Diogo and André's families, friends and all those whose lives were touched by them.
Forever in our hearts, forever our number 20 ❤️
@gregisenberg 1. Neither
2. Anthropic by a nose; then OpenAI; Google distant third
3. Tied
4. Cloud for now; local long term
5. Bootstrap til the wheels fall off, then raise if you have to
6. Best
7. Job growth for ai-enabled workforce; hard times for luddites
PM: "how long will that take? what's a realistic deadline?"
Me: "2-3 weeks, maybe 4 depending on revisions and speed of feedback."
PM: "Client wants it tomorrow, so we'll push back on them to deliver it the day after tomorrow. You're welcome."
Um. Thanks.
Well, @anthropic appears to have walked back the Claudepocalypse: `claude -p` isn't getting metered as API usage today, with no word on if/when it ever will. You have to assume it will at some point, but I can't imagine they'd risk any more negative perception pre-IPO. The only thing the threat seems to have actually accomplished: everyone burned a lot of hours and tokens figuring out how to not use Claude, myself included. I flipped to `codex exec` this morning, spent half a day talking to an agent with zero personality, and flipped right back when the news dropped.
Every agency says they're integrating AI. What that usually means is 3 people have ChatGPT Plus and 1 of them made a custom GPT that nobody uses. The gap between the "AI-enabled" claim and actually having the infrastructure to be "AI native" is enormous.
I would pay 5-10x more for claude max if regular claude, cowork and claude code were all combined and could share context with each other. having things split across all these makes me sad.