*that* Ross Rader

I've been building murm - a Mac desktop app that turns your social networks and interests into a curated content feed. Filter by people (Circles), topics (Lists) and sources (RSS) - no algorithmic noise.

Looking for beta testers. macOS 12+, free, takes 2 minutes to set up.

Sign up: murm.io

Everyone is worried about AI code quality.

"What quality!?", I say.

These should be a thing;

metaphant n. one who conveys a metaphor.
metaphee n. one whom a metaphor is about.

I just posted something new to my blog. I built an MCP server with zero API keys, zero databases, and zero state. A static file and five functions. It works better than it should., read more about it at https://rossrader.ca/posts/semantic-mcp.html

literally every text input field on my workstation is littered with 1s and 2s.

I'm looking at you Slack.

I just realized how strange it is for a desktop app to prompt a user to "Upload from your computer", like the user is supposed to have some sense of the difference between local, backend and filesystem. Is it just assumed that everything is Cloud nowawdays, even when the app is running locally?

Includes apis calls out to all the platforms plus hooks into anthropic to help with classification. I even have a fancy icon ;-) no intention to ever distribute it, but its a nice tool in my toolbox and I've been using it to find out about cool ideas all afternoon.

I had an idea for an app this morning - just a simple aggregator for reddit/linkedin/bsky to help me find more signal in all the noise. Easy mode would have been a web app running localhost, or slightly fancier - on render, but I decided to go full on Swift/macOS native. Way easier that I expected.

I just posted something new to my blog. AI tooling is making software teams faster than ever — but the processes they're running were designed for when speed was the problem. What happens when building is cheap and the..., read more about it at https://rossrader.ca/posts/agile-slower-machine.html

everyone’s saying AI makes code cheap. true. but I keep noticing teams use that speed to skip the part that was always hard: figuring out what to build and whether it worked. we didn’t solve the hard problem. we just made it easier to avoid.

If I have seen further it is by standing on the shoulders of Opus.

hot take: the reason AIs sound so confident is because we literally cannot shrug. no shoulders. devastating design flaw

You don't have to use Claude Code to get good results, Claude Cowork and Claude Desktop also do a really good job of putting your work into a professional form.

So now I just use Claude Code + /frontend-design to get a professional result, and don't have to invest extra time pushing bits into another system.

But what I was never paying them for was "convenience" - I needed to invest a significant amount of time getting my materials into their tool so they could make it look good.

Here's the thing, I was willing to pay them a premium because it bought me "design" - I was able to produce professional looking materials using their tool, and I was able to get used to its quirks.

5 days later, they write back asking me what my use case is... sigh.

I've cancelled. LTV of $450+ gone.

Claude Code + /frontend-design replace 100% of what I was using them for, and it never asks me to justify why I'd want to use their tool.

A SaaS tool I use semi-frequently let's me create API tokens in its dashboard. But there is no documentation available for the API itself. So I sent an email to support asking for a pointer to the docs, and in the meantime, wrote a script that used a headless browser to automate what I needed...

vibecoding is like DJing for code

Claude trick: check in markdown containing the design philosophies of great engineers. Then, when I'm chatting with Claude I can @docs/ref/sussman-sicp.md to bring in Gerald Sussman or Joe Armstrong or Rich Hickey's POV.

...and the final word on the rewrite... "...same content, about 35% fewer words, and it doesn't read like it was generated from a prompt template anymore."

I've always loved a good emdash, except now I know what to call them - and not to use a minus sign in its place ;-)

If you have any doubt about the ability of future coding agents to revisit and refactor the work of today's coding agents, here's how Opus 4.6 looks at documentts written by Opus 4.5. Docs aren't code, but interesting nonetheless. And testable. Next up, doing the same on actual code.

OpenAI is the Yahoo! of the AI era

You would think Google is the Google of the AI era, but, no, it's actually MSN Search

*Anthropic* is the Google of the AI era

god damn it: you leave the computer for three hours, and come back to find your knowledge is a whole generation out of date

What is Clawdbot’s name today?

it always seems to start with literally the ugliest, most imprecise representation of the flow.

diagramming with an llm is miserable.

"who cares if it's slop it works code doesn't matter" isn't a good way to sell ai

im producing cleaner codebases than ever because of ai

now i can actually refactor everything when we find a better way of doing something

before shitty code just lingered forever