Entering the new normal.
Waking up everyday to a collection of models in harnesses executing skills proactively and reactively. Your job is dreaming, managing, steering, measuring.
The adage of falling in love with the problem, not the solution has never been more powerful.
i'm obsessed with AI DIY projects
my favorite one right now is this guy who built an AI system that listens for birds outside his apartment, figures out what species they are, and paints them on his wall.
here's how the whole thing works:
1. a cheap usb mic on his balcony listens for birdsong 24/7
2. BirdNET, Cornell's AI model trained on 6,000+ species, names each bird species from the sound alone (no camera needed)
3. every time it hears one, Gemini 2.5 Flash Image paints that exact bird in the style of an Edo-period japanese woodblock print
4. the new painting drops into a live collage of everything that's been singing outside in the last 24 hours
5. and it all shows up on a framed e-ink display on his wall that reads "heard today" like a little museum placard for his neighborhood
knowing which birds visit you used to take a field guide, a trained ear, plus years of patient practice.
teddy just glances at the frame on his wall and sees the cardinal came back this morning
honestly i'm highly tempted to build one myself haha
Interested to hear your specific design thoughts. When I use GPT5.x for design it's usually terrible compared to Opus or sonnet. Plus, claude design is blowing my mind. How can I get gpt to give better design out of the box?
My workflow is Claude for product management and design and delegating to codex depending on token allowances - I've got max plans for both.
My revelation this week is how good the models I use are at product management! They (Opus, DeepSeek, GLM, GPT) have all really leveled up the game. I've been using the "grill me" skill from @mattpocockuk and the level of questions I am getting from GPT 5.5 shows a shocking depth of product knowledge and insight that it is purposing to my somewhat oddball project.
The grill-me skill is fairly concise:
----
Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.
Ask the questions one at a time, waiting for feedback on each question before continuing. Asking multiple questions at once is bewildering.
If a question can be answered by exploring the codebase, explore the codebase instead.
-----
This just worked me for an hour really honing the product so that codex could go off and
-----
Go watch his videos at https://t.co/c4TqyMoxqL - he's doing great work.
An opensource (free!) model named GLM 5.2 dropped this week and is available at all the usual inference providers. It's really, really nice. I'm using it in the opencode harness with minimal extras and it's JUST ASTONISHING how much power is available for... free! 🤓
Super easy to obsess on models given how quickly everything is moving right now. You know what? I say lets obsess for a while. 🤠
Expensive vs. Cheap
Subscriptions vs. API
Cloud vs. Local
And then you can obsess on hardware for local models, and which local models to run.
Generally I'm here to talk about building things not politics, etc. You can get enough of that easily.
Yet, in this era of the "politicization of everything" occasionally things must be mentioned.
If people want to argue about reflecting pools and algae, etc. Let them. We are too busy.
This, unfortunately seems to be all too true. With the restriction of Fable we have taken a step back technology-wise and a leap forward into government interference in our lives. We are not happy about this.
Anthropic has pushed AI forward dramatically over the past two years. It's currently the crown jewel of US AI tech.
The Feds don't like @DarioAmodei because he won't do all their bidding. And so, we've now entering the Soviet-style propaganda portion of the program with the White House feeding every reporter it can find with laughable claims like Dario is unreachable at a wellness retreat. Come on.
I'd hoped the US would not be self-defeating on AI, since it's kinda one of the last hopes the US has versus China. But here we are . . . . already
Built this iOS app on Friday that syncs to Apple Health to track workouts, supplements, etc. Took my stack about 2 hours, mostly due to my erratic product management changing things afterwards. Loads of features and does an AMAZING job of deducing macros from blurry pictures of food.
I have 3 subscriptions: Claude Max, Codex Max and OpenCode Go. These add up to 410 a month.
For $410 a month you can now have the technical capabilities of teams that have cost me $60,000 to $100,000 a month in the past.
What are you waiting for?
You can just wake up on Sunday morning and build just about whatever you can think of for practically pocket change. This time last year this was not possible.