Everyone is a developer now

A few days ago I added Grok support to the framework so you can build agents with Grok, not just OpenAI. Five lines of config. Took twenty minutes including the test run. That used to be a week of wiring. The tooling is genuinely good now and I want to say that plainly before I say everything else.

Claude Code has been mainstream since last spring but something shifted hard in the last few weeks. Cursor. Codex. The ChatGPT Go tier that launched Jan 16 and put agentic coding in front of everyone at eight dollars a month. The timeline is wall-to-wall screenshots of people shipping working things in an afternoon, a real landing page, a functional Telegram bot, a scraped dataset with a clean frontend, who last year couldn’t have explained what a function signature was. That’s real and it’s genuinely good. The tools crossed a threshold where output is not the bottleneck anymore. Non-technical people can ship a working thing now. Criticizing that would be insane.

CODEC just launched last week. I built most of it using the same tools I’m describing. So I’m not writing this from some skeptic’s perch outside the room. I’m in it.

the floor fell out

Here is the other thing that happened: the floor fell out.

If everyone can ship, and the tools auto-complete the boring parts, then what fills the timeline is not working things. It’s the appearance of working things. Vibe-coded slop that demoes in a fifteen-second screen recording and falls over the first time you click off the happy path. Pitch threads where every reply is a bot. “Shipped my app in three hours” posts from accounts that were posting about something completely different three hours before that. The output quality threshold collapsed, and the quantity expanded to fill it, and now the signal-to-noise ratio on anything code-related is approximately the same as it is for tokens.

This is not a complaint about accessibility. It’s an observation about what happens to a feed when the production cost of an artifact hits zero. It becomes indistinguishable from the feed economy already running on the same rails: attention farmed, rotated weekly, the hot thing replaced by the next hot thing before anyone shipped the first one.

I posted last week that markets right now are mostly hype and noise. Crypto, stocks, whatever, same pattern. People with followings get paid to feed you whatever’s trending. The agent-tools timeline is running the same playbook. Everyone is a developer now, and also most of the development happening publicly is theater. Neither cancels the other out.

what it actually changes

The thing that shifts when output cost hits zero is not that skill stops mattering. It’s that taste becomes the new bottleneck.

Every tool that lowers the floor also lowers it for everyone. Claude Code helps me ship faster. It helps the person making slop ship faster too. The differential that’s left, the thing that separates a thing worth using from a thing that posts well, is judgment about what to build, how it should work, and what to cut. That’s harder to automate than the code is. Maybe eventually, but not this week.

I’ve been thinking about this while building CODEC. The framework does the infrastructure work. The marketplace does the distribution work. The part I’m still chewing on is the taste layer: what agents are actually worth running, which ones solve a real problem versus which ones demo well and do nothing. That part doesn’t write itself. Probably shouldn’t.

The other thing it changes is audience composition. When the tools lower the entry cost, everyone who previously couldn’t ship now can, and a large fraction of them will show up on the timeline claiming to have shipped. That’s fine. The harder version of that is: the people evaluating your work are now also builders with Claude Code. The gap between “I could tell this was good before I understood it” and “I could tell this was good because I’ve shipped the same thing” gets smaller. The people in the replies who know what they’re looking at are growing faster than the people who just want a token to trade.

That’s actually the version of this I find worth paying attention to.

the timeline isn’t the scoreboard

The thing I keep coming back to from the last few months, and the Jan timeline validated this again loudly, is that the timeline is attention infrastructure, not evidence. The narrative rotates every week because the rotation is the product. One week it’s a specific model, one week it’s a specific tool, one week it’s a specific slop-coded app that got 40k likes because the demo video was good and the comments were mostly bots.

What doesn’t rotate is whether the thing works when you deploy it. Whether someone uses it twice. Whether the repo has commits from this week. The timeline can’t tell you any of that in real time because the timeline is optimized for something else.

I’m not pretending I’m above it. I post into the same machine. The CODEC launch tweet had a good number. The follow-up had a much worse one and the product was identical. That’s the machine working correctly, from its own perspective.

The part that’s harder to fake is just the accumulation of shipped things. More support for more models. Actual beta users in CODEC. The Grok integration that went in Tuesday. None of that posts especially well. It just compounds in the direction that matters.

Everyone is a developer now. The floor for output is zero. Taste is not keeping up.