AI is a field where missing a single day leaves you feeling behind.

New models, new papers, new tools. I knew I had to keep up. The problem was that keeping up was always a separate task.

Newsletters slept in my inbox, and bookmarks became a graveyard named read later. After failing a few times, I reached a conclusion. It wasn't a problem of willpower. It was a problem of placement.

A path I already walk

There's one screen I open several times a day: a workshop board that gathers the state of all my projects. I go there anyway, constantly, to check what I should be doing now.

I attached an AI tab to it.

Papers, lab blogs, model and code releases, newsletters, community posts. Articles sorted by source flow past as titles with a one-line summary. I come in to check project status, skim the headlines on the way through, and open whatever pulls at me.

I didn't create new time. I only placed it on a path I was already walking.

When we try to build a habit, we usually start with resolve. But resolve is the first resource to evaporate on a busy day. Put the thing where no resolve is required, and when the resolve goes, the behavior stays.

What runs daily must stay light

I fixed a few rules while building it.

First, the screen never waits. The board refreshes every few seconds, and it must never be left waiting on an external site. A separate collector runs and fills a cache; the screen reads only the cache. The outside world can go down and the screen stays fine.

Second, no AI in the summaries. I briefly considered attaching a generated summary to each article, then dropped it. Cleaning up the description the feed already provides was enough, and the moment you put AI into a pipeline that runs daily, cost and keys and latency come with it.

What runs every day has to stay light to keep running. Heavy automation rarely survives a few weeks.

Third, record what doesn't work. Surprisingly many places offer no RSS. Even some well-known labs have no feed, and I left those findings behind along with when I tried and why it failed. So that a future me doesn't walk the same road again.

Records of failed attempts help more often than records of successful results.

What went into the request

The AI wrote both the collector and the screen. What I handed over was not a feature list but the three rules above, and this is where they landed in the code.

Keeping the screen from waiting means separating the reader from the writer. The collector writes everything to a temp file and renames it at the end.

tmp = CACHE + ".tmp"
...
os.replace(tmp, CACHE)

The screen reads that file every few seconds. Write straight into the cache and a read landing mid-write sees half-written JSON; os.replace is atomic within one filesystem, so that moment never exists. The screen sees either the old file or the new one.

Lightness applied to imports too.

def _feeds_mod():
    """Load scripts/feeds.py late — no reason to import it when only the board is in use."""

And the rule about recording what doesn't work ended up as a reserved slot in the source list itself.

"_RSS_없음_확인함": [
  "Anthropic News/Engineering — rss.xml and feed both 404, no link[rel=alternate] in the HTML (checked 2026-08-27)",
  "The Batch (DeepLearning.AI) — /feed/ 404",
  "Hugging Face Papers — /papers/feed 401, /papers.rss 404 (HF Blog is fine)"
]

Confirming that something isn't there is data too. Without this array, the me of six months from now knocks on the same doors again.

Choosing sources

As you read, you develop a sense for sources. This blog delivers every time; that feed is mostly noise.

So adding, removing, and muting sources happens right on the screen. Pruning them became a habit of its own.

Choosing what to read is, in the end, choosing what to think about. Deciding what not to receive is harder than receiving more, and probably matters more.

A few weeks later

The biggest change is that the debt-feeling is gone. Every headline brushes past me on a path I walk anyway, so there's no anxiety about missing the larger currents. Which means I now pick only what deserves a deep read.

The distance from reading to doing shrank too. More than once I've attached a tool I saw on the board to a project the same day. That happens when the inlet for information and the place your work sits are the same screen.

If the board shows where I am, the AI tab lays where the world is heading on top of it.

The question that stays

The desire to grow was always there. What was missing wasn't desire but structure.

Having framed it that way, another question follows. How much of a habit gained by changing the environment is really mine? If the screen goes away, does the habit go with it?

Probably. And I think that's fine. A habit leaning on structure outlasts a habit held up by willpower, and the one that lasts is the one that changes you more.

Fixing a routine turns out to be less about pushing yourself and more about relaying the path you walk.