EPISODE 06 · Track AI trends · 2026.05.17

Chasing headlines
isn't understanding trends.

This is a talk about keeping up with AI.
More accurately — it's a talk about not mistaking "scrolling more" for "understanding more."

Start with the word →
AGENT 101 SERIES · EP 06 · NO-BULLSHIT EDITION
§ 01 · Start with the word

"Headlines," "information," and "trends" are three different things

Hundreds of AI headlines hit you daily. Answer one question: the last "bombshell" you scrolled past — did it actually change something you do? This is where most people are stuck — mistaking "scrolled it" for "understood it," and "volume of information" for "judgment."

  • CONCEPT · 01 · HEADLINES

    Chasing headlines = getting pulled by the feed

    Headlines are fed to you by an algorithm whose goal is to keep you scrolling, not to make you smarter. Headline-chasers get dragged by the feed — today's "disruption," tomorrow's "game-changer" — looking busy while donating watch time to a platform. The more you scroll, the more anxious you get, and your judgment hasn't moved.

  • CONCEPT · 02 · INFORMATION

    Information = raw material, mostly noise

    Every headline, every release, every hot take is raw material. But material isn't knowledge — 95% of it is noise: irrelevant to what you do, or overturned by tomorrow. Information has no value in itself; it gains value only after you filter it and combine it with your judgment.

  • CONCEPT · 03 · TRENDS

    Trends = a direction filtered from noise + your judgment

    The only use that loses information when deleted. A trend isn't "what just happened," it's "the direction that recurs and will affect what you do." It requires you to filter signal from a pile of noise, then judge against your own situation. People who understand trends read less, but everything they read is usable.

§ 02 · Audit your intake

"I read AI news every day" is an empty phrase

It sounds like diligence, like ambition. But it fails the three simplest checks. If it fails, what you're doing isn't "keeping up with trends" — it's scrolling to relieve the anxiety of missing out, anxiety rising, judgment standing still.

  • INTAKE TEST · 01

    Did what you scrolled actually change a decision?

    Not "I learned about this news," but "because of this, I changed a practice / a judgment." If you can't name one decision it changed, this scrolling is pure consumption — no different from scrolling short videos.

  • INTAKE TEST · 02

    Can you say on the spot whether a piece is "signal" or "noise"?

    Signal affects what you do; noise looks exciting but is irrelevant to you. If every piece makes your chest tighten yet you can't tell whether to care, you have no filter — you're paying with attention for all information indiscriminately.

  • INTAKE TEST · 03

    If you stopped scrolling for a week, would your judgment get worse?

    The brutal one. If pausing for a week leaves your decisions unaffected, you weren't scrolling trends — you were scrolling FOMO. Real signal, missed for a week, you'd genuinely feel behind; pure noise, dropping it only makes you clearer.

The first is FOMO — endless to miss, impossible to catch. The second has a scope, a cadence, and a failure signal. Understanding trends isn't seeing more — it's missing the right things: deliberately drop 95% of the noise to see the 5% of signal.

§ 03 · After the word

"Pulled by the feed" and "scanning with a radar" are two completely different intakes

Two people both want to keep up with AI; a year later their judgment is a tier apart. The difference isn't who scrolled more — it's whether their head runs "dragged by the feed" or "actively scanning with a radar." This is the only contrast you need.

PULLED BY THE FEED · what you might be doing

More scrolling, more panic

Dragged by the algorithm and trending lists, wanting to see everything, hoarding a pile never read.

  • Open and scroll, pushed along by the feed
  • Chase every field, go deep in none
  • Hoard "read later," never read
  • Anxiety keeps rising, judgment doesn't improve
SCANNING WITH A RADAR · who you're becoming

See less, understand more

A few high-signal sources, only what's relevant to you, one brief instead of scrolling.

  • Fixed 3 areas + a few primary sources
  • Ask one thing: "what does this change for me?"
  • One weekly brief replaces daily scrolling
  • Anxiety down, judgment updates steadily
§ 04 · How to build the radar

Four steps: Sources → Cadence → Capture → Synthesize

A working trend radar is these four steps turning. Most people only do "look" (and passively at that) — no source selection, no cadence, no synthesis — so they scroll forever and stay anxious. Remember the four and you'll know exactly where your information system breaks.

RADAR LOOP SOURCES CADENCE CAPTURE SYNTHESIZE
  • 01

    SourcesSOURCES

    Pick a few high signal-to-noise sources, in three tiers: primary (lab blogs / release notes / papers), community (HN / GitHub Trending), people (a few researchers and builders who actually ship). Source quality sets the ceiling on your radar.

  • 02

    CadenceCADENCE

    Set a fixed frequency: once a week (or a daily brief), not constant scrolling. Cadence turns "passively pushed" into "actively fetched" — the key to quitting FOMO. Real signal doesn't vanish because you saw it a week late.

  • 03

    CaptureCAPTURE

    Let the system fetch automatically; don't rely on "remembering to look." RSS, subscriptions, a scheduled agent — wire sources into a pipe that runs itself. A radar that depends on willpower decays; automated capture is what holds.

  • 04

    SynthesizeSYNTHESIZE

    Have an agent compile the fetched raw material into one brief: dedupe, rank, flag "why it matters." You only read conclusions and judge. This step turns "a pile of information" into "a few signals" — and it's where a trend radar truly saves your time.

§ 05 · The source map

Sorted by "signal tier," not by hype

Trending lists sort by "hottest" — useless for judging trends. This map sorts by "how close to primary" into four tiers: higher is closer to the source, with less noise. The principle: go upstream when you can, don't only read second-hand takes.

  • TIER · 01 · PRIMARY SIGNAL second-hand take → the source

    Read the source when you can: see it before others paraphrase it

    Contrarian take: 99% of people read "an interpretation of an interpretation of the source," distorted at each layer. Real signal is upstream — official lab blogs, release notes, technical reports, papers (arXiv). Read the source directly and you're a step earlier and a hand more accurate than second-hand readers.

    SOURCESprimary signal
    LABSofficial blogs / announcements from major AI labs
    RELEASEproduct release notes / changelogs / docs
    PAPERSarXiv (cs.AI / cs.CL) · abstract + conclusion is enough
    RAW DATAofficial benchmarks / leaderboards · read numbers, not headlines
    PRINCIPLEif you can read the source, don't read the paraphrase
    lab blogs release notes arXiv official docs
  • TIER · 02 · COMMUNITY SIGNAL influencer shouting → builders using

    Watch what builders use, not who's shouting

    Contrarian take: marketing accounts tell you "what's hot"; communities tell you "what people actually use." Hacker News discussion, GitHub Trending stars, specific technical communities — these are where engineers vote with their feet. Whether a tool has real value: watch whether builders adopt it, far more reliable than watching the hype.

    SOURCEScommunity signal
    HNHacker News · read the top skeptical comments, not just titles
    GITHUBGitHub Trending · who's gaining stars · real issue feedback
    COMMUNITYspecific technical subreddits / Discords · real usage
    SIGNAL"builders migrating to X" beats "X raised $100M"
    PRINCIPLEwatch votes by feet, not by mouth
    Hacker News GitHub Trending technical communities
  • TIER · 03 · PEOPLE / CURATORS follow influencers → follow builders

    Follow a few people who ship, not the loudest

    Contrarian take: volume and signal-to-noise are often inversely related. The ones posting "AI disrupts everything" daily are mostly selling anxiety. Find a few high signal-to-noise people — those actually doing research or building products. They post less, but every post carries information. Following the doers lets them do your first layer of filtering.

    SOURCEShow to pick people
    KEEPpeople doing research / building · posts with detail and code
    KEEPpeople who'll say "I'm not sure / this is overhyped"
    DROPdaily "mind-blowing / disruptive / learn now or lose"
    DROPpure paraphrasers who never build, growing on FOMO
    PRINCIPLEsignal-to-noise > follower count
    researchers builders high signal-to-noise
  • TIER · 04 · NOISE ZONE · MUTE passive intake → active blocking

    Actively muting this tier matters more than adding another

    Contrarian take: building a radar isn't only "add good sources," it's "actively delete bad ones." Clickbait digests, FOMO accounts, second-hand takes, "AI trend mega-predictions" — these drain your attention without updating your judgment. Muting / unsubscribing them raises your signal-to-noise more than adding three new sources.

    MUTEactive block list
    UNSUB"50 AI tools a day" FOMO digests
    MUTEaccounts whose titles are always "insane / disruptive"
    SKIPsecond- and third-hand interpretations of the source
    OFFall "algorithmically recommended" AI feed notifications
    PRINCIPLEdeleting noise beats adding signal
    unsub FOMO digests kill notifications skip second-hand takes
§ 06 · Run it tonight

Three steps — stand up your trend radar tonight

Understanding "build a radar" isn't enough; you need one actually running. The three below you can start tonight — subtract first (delete noise), then automate (build a brief), then practice judgment (vet a claim). Don't aim for perfect; first get the radar turning.

DEMO · 01 15 min · one-time

Subtract first: set 3 areas + 3 primary sources, delete every FOMO source

The first step of a radar isn't adding, it's deleting. Tonight, figure out the 3 areas you truly care about, pick 1 primary source for each, then unsubscribe / mute every FOMO digest and algorithmic feed. Once you subtract, your signal-to-noise doubles.

  • Write down the 3 areas you actually care about (don't be greedy)
  • Pick 1 primary source per area (official blog / docs / an arXiv category)
  • Unsubscribe from every "N items a day" FOMO digest
  • Turn off "AI recommended" feed notifications in social apps
  • Keep only these 3 sources — actively give up the rest
my trend radar · source list
Area 1Agents / tools → lab blogs + release notes
Area 2my industry × AI → 1 high signal-to-noise person
Area 3local models → GitHub Trending + HN
— unsubscribed 6 FOMO digests, killed 3 notifications —
✓ noise halved · radar in place
DEMO · 02 30 min setup · then automatic

Build a weekly AI brief agent

Swap "scrolling" for "reading a brief." Have an agent, at a fixed weekly time, fetch the important developments in your 3 areas, dedupe, rank, compile into one brief, and push it to you. You only read conclusions. (This is exactly what this site's daily news pipeline does.)

  • Use Perplexity / Claude + a scheduled task (or n8n)
  • Write a brief spec: 3 areas, only "changes a practice" developments, with sources
  • Set it to run Sunday morning, output markdown
  • Push the result to email / Slack / Notion
  • After reading, ask: which item this week is worth changing a practice for?
weekly-ai-brief · Sundays 08:00
01FETCHpull primary sources for 3 areas (last 7 days)
02FILTERdedupe + drop the irrelevant · flag "why it matters"
03RANKorder by "will it change what you do"
04BRIEFcompile a 1-page brief + source links → email
✓ 5 signals · primary sources attached · scrolling at zero
DEMO · 03 per claim · 3 min

Practice judgment: have AI "vet a claim" for you

See a "bombshell"? Don't rush to believe or reshare. Build one habit: have AI take you back to the primary source, give the counter-case, and flag the exaggeration. Three minutes, and you can tell signal from marketing.

  • Paste the claim / headline to Claude or Perplexity
  • Have it find the primary source and check what it actually says
  • Have it give the "this is overstated / the opposing angle"
  • Ask "what does this actually mean for someone doing X"
  • Three questions answered, signal vs noise is obvious
claude · vet a claim
YOU"model X crushes everything" — take me back to the primary source
CLAUDEchecking the original report…
▌ Primary source says: leads only on one narrow benchmark
· headline "crushes everything" = overstated
· actual impact on you (doing X): little for now
✓ verdict: marketing noise · swipe away
§ 07 · The next 90 days

From chasing headlines to understanding trends

This 90-day path doesn't promise you'll "miss nothing." It gives four specific actions, each with an explicit failure signal. If a week's signal lights up, go back to the previous week. Don't push forward.

  • WEEK 1 – 2 · Cut FOMO first

    Unsubscribe from every FOMO source, keep only 3 primary ones

    Subtract: unsubscribe from all "N items a day" digests, kill algorithmic notifications, keep 1 primary source for each of your 3 real areas. Failure signal: by end of week 2 you're back to scrolling the feed. What to do: revoke those apps' notification permissions outright — cut the entry point.

  • WEEK 3 – 4 · Set the cadence

    Fixed once a week, not constant scrolling

    Change "scroll anytime" to "check once at a fixed weekly time." Don't touch AI feeds otherwise. Failure signal: you still can't resist opening it several times a day. What to do: remind yourself "signal doesn't vanish if seen a week late" — real signal survives the wait.

  • WEEK 5 – 8 · Automate capture

    Build a weekly brief agent, let synthesis run itself

    Use an agent + scheduled task to auto-fetch developments in your 3 areas and compile a brief, pushed weekly. Failure signal: the brief exists but you still can't resist manual scrolling. What to do: go back to weeks 1–2 — the FOMO sources aren't fully cut, and tools can't save a habit.

  • WEEK 9 – 12 · Wire into judgment

    Ask of every signal: "what does this change for me?"

    Wire the brief into real decisions: for each signal, force yourself to answer "what practice does it ask me to change"; can't answer, swipe away. Failure signal: by week 12 your brief has never changed a decision. What to do: your areas are too far from what you do — go re-pick the 3.

After 90 days you should be able to answer: "In the past month, which AI development actually made me change a specific practice?" — if you can answer, pointing to the change, you've graduated from "chasing headlines" to "understanding trends." If you can't, you spent 90 days scrolling more news while your judgment stood still.

Real Agent Use Cases

A hundred headlines are worth less than one signal that changes a practice

The headlines you scroll are forgotten by tomorrow. But a radar wired into your judgment catches the signal and blocks the noise when it matters.
"Execution is always undervalued." If you don't delete the FOMO sources and stand up a first radar tonight, tomorrow you're still dragged by the feed. That's not a pep talk. It's an observation.

← EP 05 · Talk to AI　·　EP 06 (you are here)　·　Agent 101 · all 6 episodes complete