2026 is the year we will stop hand-writing code and fully mechanize software engineering.
This might sound like hype. And certainly thereโs a lot of hype going around! But something substantial lies underneath the bluster.
Hereโs how Iโm thinking about it.
A tip: any time you ask Codex to complete a task, just slap /goal in front of it.
Far more persistence and looking at it from different angles to ensure it completed it.
I think it's important that we continue taking a moment occasionally to remember that MACHINES CAN WRITE NOVEL CODE AND IT ACTUALLY MOSTLY WORKS lest we forget how amazing this is
@steveruizok@schickling@kostyafarber_ also one of the biggest gaps IME is the agent isn't great at architecture diagrams
could definitely use some skills, probably with real examples
@steveruizok@schickling@kostyafarber_ โค๏ธโค๏ธโค๏ธ I cannot wait for a desktop app!
Working with an agent in a session while you can both access the canvas is magical.
Have you implemented some version of "watching" the canvas for changes? (channels in MCP can do this)
avoids having to say "look at my changes"
@appfactory@wesbos I'm with you!
And yet, I also think you want agents IN your software too (they see everything you see, they can touch everything you can touch).
But chat. It's mostly just chat.
https://t.co/eDc9KeJTbS
The definitive UI paradigm for AI has already been established.
It's Chat.
It's unfortunate. But thems the Facts.
Just like you can't reinvent the steering wheel when designing a car, you can't escape Chat.
I have a particularly challenging feature to build today in Alexandria: using a localhost web UI as an shared agentic interface with a Claude/Codex CLI session.
Going to do a live-dev thread to try to trick myself into building it.
Wish me luck! ๐