"Rent the model, own the system" is the cleanest version of this I've seen, the integration and eval and observability layer is the part that actually compounds.
@whats_ai
-
Per-task routing becomes default, single model apps feel dated
By
–
Per-task routing is becoming the default, locking your whole app to one model already feels dated.
-
Self-Prompting Systems: Beyond the Prompt Itself
By
–
"Build a system that prompts itself" really lands, writing the prompt was never the hard part, it's everything you wrap around it.
-
Midjourney uses diffusion, Nano Banana uses autoregressive image generation
By
–
Do you know how Midjourney and Nano Banana actually create images?
Completely differently! Midjourney uses diffusion:
It starts from pure noise and sharpens it a little at every step, like sculpting. Nano Banana is autoregressive:
It writes the image token by token, exactly -
Three ways to run sub-agents: Fork, Teammate, Worktree
By
–
Nowadays everyone think sub-agents are magic.
They're not. There are basically 3 ways people run them right now. 𝟭. 𝗙𝗼𝗿𝗸
The agent spins off a task, gets the result, and continues. 𝟮. 𝗧𝗲𝗮𝗺𝗺𝗮𝘁𝗲
Multiple agents work together and share context. 𝟯. 𝗪𝗼𝗿𝗸𝘁𝗿𝗲𝗲 -
Agents writing own smoke-tests builds trust in output
By
–
The agents writing their own smoke-tests is the first habit that actually makes me trust the output more haha.
-
Abstraction layer is plain markdown for both human and model
By
–
Funny how the abstraction layer ended up being plain markdown the human and the model both read.
-
Open-weight model sits close to the frontier
By
–
An open-weight model sitting this close to the frontier is something!
-
Open weights: irrevocable access beats benchmark points
By
–
Strongest case for open weights I've seen lately, access no one can revoke beats a few benchmark points.
-
The collage theory vs the truth about AI image generators
By
–
A friend asked me last week if AI image generators "steal" pieces of existing images and glue them together. Before replying, I thought about why he believes that.
And honestly, the collage theory sounds more believable than the truth. The truth is, models learn before they