An interesting thing about the AI era is you no don't need humans to figure out "how" or "what". New to a codebase and need to trace a code path or discover relation between components? Ask an agent.
Now most questions to my teammates are about rationale or judgment - the "why".
Wave was open-sourced, no? Became an Apache Location protocol iirc
Google Wave is a pretty perfect conceptual interface for AI + multiplayer human collaboration.
Find someone who loves you as much as Gemini loves filling in `gemini-1.5-pro` as the model name for everything.
This is the most resonant definition of slop I've seen: "It’s when we prioritize volume over impact."
Because this accurately captures the things that are slop but not AI (SEO farms, bad genre fiction, a whole ton of kid YouTube...).
Is it weird to say that coding with agents feels more...organic?
It feels more like tending a garden than building a house. You plant seeds (write specs), water them (poke it along), pull weeds (refactor messes).
It's a different skillset but very informed by existing knowledge.
With Cloudflare forking Next.js and tldraw yoinking unit tests from their public repo, it seems like something significant is happening in open-source this week. We're entering the age of the Slop Fork.
Who implements this today? Nobody. But if you're a library author, it couldn't hurt to start adding this. If you're an agent builder, it couldn't hurt to start detecting it. Let's make it easier to teach agents how to use open source!
.agents.lib is a directory distributed with open source packages that contains guidance and configuration for common coding agent customizations (specifically AGENTS.md, skills, and MCP servers).
Rather than invent new distribution mechanisms, let's use package managers!
If you're a library author, it's a pain to figure publish agent skills in a way that your users will actually see.
If you're a developer using libs, it's a pain to make sure you're using the best and latest agent guidance for them.
Let's solve it! I'm proposing .agents.lib
Engineers dunking on Claude Code's memory usage or perf are really telling on themselves.
Claude Code went from zero to $2.5B run rate business in a year. Every engineer should aspire to such colossal success and impact of their work. You can optimize later. Win first.
I use this prompt far more often than I'd like to admit:
"I've done a bunch of work without properly breaking it into small commits. Please look at the current state and do some incremental commits. Each commit must have tests passing."
That being said, it works pretty well!
Has there ever been a successful company that launched itself by buying an incredibly expensive domain name or a superbowl ad? Because all I can think of are abject failures and it's surprising the lesson doesn't get learned by either founders or their investors.
It feels like a glaring oversight that, despite inventing MCP, Anthropic did not work to incorporate either hooks or skills into MCP when it would obviously fit, help standardize, and solve a bunch of problems.
For some reason I can't stand tab completion in AI coding. If I'm actually touching code these days, it's because I have specific stuff I want to do and I need language server autocompletions not LLM guessing.
Companies with eccentric bespoke build systems have shot themselves in the foot in the AI era since these types of problems in particular hamper the "write, check, fix" inner loop that agents need to successfully build software.
The difference between "too early" and "too late" in the AI era is ~2 months.
I'm not sure any TV series will ever have as ruinous of ending as Game of Thrones. This is largely because few shows will ever be as good as it was at its peak.
It took really special circumstances to be that good and then fumble that hard.
TIL that XXL and 2XL are in fact not the same size and I'm still reeling.
The real way that the ultra-rich avoid paying their share of taxes is by being able to borrow cash at extremely low interest rates secured by illiquid assets, thus being able to realize the benefits of liquidity without triggering capital gains. Fix that, don't do wealth taxes.
My contrarian awards season take is that 28 Years Later deserves to win "Best Original Song" for "Boots". It's grating and unpleasant but one of the most effective auditory cinematic experiences I've ever had. Truly makes my skin crawl.
Oh wait, but Sinners also deserves it...
The best outcome for AI is a no-clear-winner ecosystem where there are dozens of different agents forced to interoperate via standards like MCP.
Such a result would keep pressure and incentives towards open, hackable systems like Web 1.0 and less like app store walled gardens.
The show so far is basically just extrapolating from the answers to those two questions, and it's just incredibly compelling.
It also subverts "mystery show" tropes by setting up and then immediately revealing the answer to questions instead of having them linger. So good.
Pluribus is maybe the best high-concept writing of all time. It takes an inherently captivating premise: "What if a hivemind took over nearly all of humanity overnight?" and layers on an incredible answer to the question of "And why don't they just kill the remaining outliers?"
I'm fine with people sending me an AI-generated report and saying "I had AI research X and there was some interesting stuff". It sets expectations that I should skim for interesting bits not read carefully.
If you want to communicate, though, send the prompt not the output.
If you want to use AI to aide your writing do it by having the AI help you organize your thoughts into an outline or summarize a bunch of research. Have the AI critique your writing for brevity and clarity.
Writing is communication. Every word written should have intent.
I dislike AI writing slop much more than other forms because there's no transformation of intent, only bloat.
Media gen: transform language to visuals
Code gen: transform language to behavior
Text gen: transform language into puffed-out language
It literally just wastes time.
Adding "It's not [just] X, it's Y" to my list of things to avoid in writing forever now.
ML researchers creating SWE-Bench which only tests Python is like a hockey fan creating a hockey trivia benchmark and calling it Sports-Bench.
It's maddening that the only benchmark every lab universally hill-climbs on is so deceptively limited.
If you are creating a tool that in any way allows image input, the single most important UX polish you can add is supporting pasted images. Do it yesterday.