Hype Radar
Living record, rebuilt from the database at 2026-06-19 05:39 UTC. Not hand-edited.
What this is, and why I'm doing it
AI is moving faster than anyone can read. Most of what crosses my feed is noise; a few things each week genuinely change what I can sell, what I can ship for clients, or how I run my own shop. Hype Radar is the filter. It watches a curated set of AI operators on X, captures everything they post, and surfaces the handful that actually matter — judged against my real book of business: GIG, my clients, and Articulate.
Why I built it: so I stop missing things, and so "keeping up with AI" becomes a system instead of a panic. The job isn't to collect tools — it's to find, make sense of, decide, and act. Every item gets read through one question: does this help me, GIG, or a client do something better this week?
What I do with the database: every item worth keeping is scored P/Q/S — Personal productivity, Quality, Sellable (0–3 each, out of 9). Six or more means act on it; four to five means watch for a trigger; three or less is skipped. The keepers become things I test, things I publish (newsletter, LinkedIn), and the raw material that feeds the Better Tribe field guides. The Airtable database is the system of record behind all of it.
What rose to the top — today
Read against GIG / me / my clients. Each note says who it's for and what to do with it. ★ = you flagged it directly.
Gemma 4 (12B/26B, Apache-2.0, multimodal, runs on a 16GB laptop). The credible OPEN, LOCAL alternative to Claude — data-sovereign for UAE clients (PDPL), zero per-token cost on the Studio. Promote as the 'you don't have to send data to a US API' option. Verified (Google). Feeds Better Tribe / Compute.
CORE Articulate lane. AI-answer visibility is exactly the GIG GIGI + Fintan + GEO-offer play. Multiple panel voices flagging same week = strong signal. Route to Anselm/GEO workstream; fold into the offer + a client-facing explainer. Verified voices.
Apache-2.0, runs locally, no LLM/cloud; beats pymupdf4llm/markitdown on 3 benchmarks. One-click agent skill. Slots straight into our PDF/proposal ingestion + doc pipelines. Verified. (Was in this morning's pull — missed first time.)
YOU: house-CSS + client sites + decks. Stays on-brand across projects, /design-sync pulls the system into the repo. Adopt now — directly upgrades how we ship every client surface. Verified (Anthropic).
Conditional — worth a look
Real editor (not clip-gen) that watches footage, understands it, makes cuts; given to editors behind Key & Peele / Beast Games. Test for GIG reels. Availability unconfirmed (launch announcement). One of the 'P' video tools Anthony saw. (Was in pull — missed first time.)
[AB flag] Generation lives on the timeline — Claude/Codex trim, reorder, generate footage in one project. Test for GIG reels + Kitroom video production. YC-backed. Verified. (Missed by the radar — @MakerThrive isn't in the voice pool.)
GIG/clinic/RE visuals + Dylan's lane. Real-time iteration speeds creative. Trial vs current Higgsfield/Firefly stack. Feed-reported — confirm capability before client use.
YOU/Studio: $0 local frontier model for mechanical/sensitive volume; data-sovereign angle for UAE clients (PDPL). Test on the Studio vs current Ollama routing. Verified.
YOU: avatar presents the weekly meeting / internal explainers (your stated want) — TRY this week. GIG explainer tier (not hero reels). Clinic/RE volume content. Free = zero downside. Newsletter item. Verified.
Brand-aligned design agent — slides/social/one-pagers on a Figma-like editable canvas, brand import from URL. $7.5M seed (General Catalyst). Competitive-intel + positioning hook vs the Articulate brand-asset offer, not a tool to adopt. Verified.
New video-model versions surfaced in a body-physics comparison (Grok Imagine 1.5, Gemini Omni Flash; Seedance 2.0 & Kling already logged). Test against Seedance for GIG reels. Feed-reported — version specifics unverified.
Open-source computer-use driver; Linux backend now in pre-release (X/Wayland). Agents (Claude Code/Codex/own loop) drive native desktop apps via CLI or MCP without stealing focus. Re-score at Linux GA. Verified GitHub trycua/cua.
STORY/trend (technique, not a single tool): solo operator running a full GTM motion via agent + outbound stack. Direct Articulate offer angle ('one-person GTM engine'). From in-pull GTM-engineer threads I ignored first time.
STORY/capability. Admins provision MCP connectors org-wide via IdP (Okta) — 'central OAuth'. Matters when clients scale connectors; an Articulate deployment/onboarding angle. Verified. (Not in this morning's pull — For You slice missed it.)
[AB flag] AI/social-selling CRM angle — maps to the MEP lead-management workflow + the Christies/Paul lead-mgmt POC. Confirm exact product (likely Breakcold) before recommending. Missed by the radar — poster isn't in the voice pool.
STORY, not a tool we adopt. A dominant game engine shipping native MCP (Claude Code/Codex drive the Editor) is a strong 'MCP as universal control plane' signal that reinforces our MCP-first positioning. Verified.
YOU: Claude/agent token economy. If the 95% claim holds it cuts our biggest run cost. Investigate + benchmark before adopting — claim unverified.
The written summaries & what feeds Better Tribe
The radar is an input to things people actually read:
Tier-1 ACT-ON items route into the matching Better Tribe field guide; ≥6 items also seed a LinkedIn draft and the weekly newsletter. (Newsletter + per-field auto-write are the next build.)
Next steps
- Re-score the full Feed through the relevance lens so flagged/buried items (Palmier-class tools) surface automatically — the classifier currently dumps real tools as "noise/opinion".
- Add Reddit + Hacker News as sources — both JSON-accessible, free, no quota; they'd have caught the tools your X algorithm surfaced. Then Product Hunt + the AI newsletters.
- Finish the 126-voice pull — throttle across rate-limit windows (or lift X budget) to clear the 63 that hit the 402 cap.
- Wire build + summary into the headless job so this page and the daily summary self-refresh without Claude open.
- Build the missing outputs — auto-write to Better Tribe hype.json, the weekly newsletter, LinkedIn drafts; and close the loop with an "implemented?" field per ACT-ON item.
The story so far
How this engine got built and fixed — the working thread, kept because the reasoning is part of the record.
Under the hood
Architecture, scripts, status & recent fixes (the plumbing)
Two sources feed one scoring pipeline: a lossless per-author X API pull over the curated panel, and a 6-hourly For-You cookie capture. Items are classified into a 5-tier taxonomy, P/Q/S scored, deduped and cross-confirmed, then written to the database, the scored log, and Better Tribe.
| Script | Role |
|---|---|
scripts/foryou_to_feed.py | The engine — capture → classify → Airtable Feed (agent-free, launchd every 6h) |
scripts/foryou_capture.py | For-You cookie capture → 15-day archive |
scripts/build_page.py | Rebuilds this page from live state |
context.md | The relevance map (GIG / me / clients) the synthesis reads |
Status: daily per-author pull RUNNING (07:33) · 6-hourly engine CAPTURING · deploy via cwdeploy v4.0 (mutex).
Latest run:
## [2026-06-19 05:33 UTC] PANEL pull (126 voices, 3d) — tweets 293 | reconciliation FAIL | _akhaliq:200:12 AhmadMustafaOs:200:noid alanwhere:200:noid aleyda:200:6 amandanat:200:0 amasad:200:12 andy_matuschak:200:2 annhandley:200:0 anthilemoon:200:0 AnthropicAI:200:8 aprildunford:200:0 bcherny:200:1 becc_holland:200:noid beerud:200:0 BenPines:200:0 bernardjhuang:200:1 BionerdAdam:200:noid bmoesta:200:0 btaylor:200:0 c_valenzuela_:200:0 calnewport:200:0 CasJam:200:8 chaseleanbiz:200:noid chriswalkerCEO:200:0 cole_medin:200:0 copyhackers:200:0 corridorcrew:200:0 cpedregal:200:0 curtispyke:200:0 CyrusShepard:200:4 danshipper:200:10 david_perell:200:3 deepgramscott:200:0 destraynor:200:1 dickiebush:200:41 doist:200:0 EddieShleyner:200:noid ekuyda:200:0 emilyinseo:200:1 emollick:200:33 fabianstelzer:200:8 FadekeAdegbuyi:200:noid fortelabs:200:9 gaetano_nyc:200:9 garyvee:200:26 georgehotz:200:noid goldiechan:200:0 GoodMarketingHQ:200:0 GrammarGirl:200:0 greggunn:200:0 gtdguy:200:0 harrydry:200:0 higgsfield_ai:200:1 hosamarab:200:0 hotdogsladies:200:0 hwchase17:200:4 IndyDevDan:200:0 iPullRank:200:2 jaredrobin:200:noid jasondbay:200:0 jasonfried:200:7 JaysonBay:200:0 joaomdmoura:200:4 jspector:200:0 Julian:200:0 JustinMichael11:200:0 karenkx:200:0 karpathy:200:0 KeenanonSales:200:noid KentBeck:200:0 Kevin_Indig:200:5 khemaridh:200:2 kieranklaassen:200:9 kippbodnar:200:0 kwindla:200:4 kylemcoleman:200:noid lennysan:200:12 levelsio:200:48 lilyraynyc:402:noid LinusEkenstam:402:noid magnitt:402:noid maimedhat:402:noid mappletons:402:noid Marie_Haynes:402:noid MarketingProfs:402:noid markosoglow:402:noid matistan:402:noid MattNavarra:402:noid metehan777:402:noid MichLieben:402:noid mosseri:402:noid mvanhorn:402:noid nateliason:402:noid NealSchaffer:402:noid nickabraaham:402:noid nickabrbs:402:noid nickfloats:402:noid Nicolascole77:402:noid NikolasRuge:402:noid OfficialLoganK:402:noid p_millerd:402:noid patrickstox:402:noid paulg:402:noid paultrillo:402:noid randfish:402:noid RayDalio:402:noid retentionadam:402:noid robhardy:402:noid runpod_io:402:noid runwayml:402:noid SahilBloom:402:noid sarah_edo:402:noid sgrogelberg:402:noid simonw:402:noid StephHabif:402:noid suganthan:402:noid swyx:402:noid tferriss:402:noid theChrisDo:402:noid thejustinwelsh:402:noid tomcritchlow:402:noid tonystubblebine:402:noid trq212:402:noid vaibhavgupta_50:402:noid visakanv:402:noid WizardofAdsRoy:402:noid
Recent fixes / failures:
- FAIL-VERSION-IN-META-NOT-ON-PAGE + FAIL-DEPLOY-ALIAS-SILENT (2026-06-19)
- FAIL-HYPERADAR-6H-SILENTLY-DEAD-SINCE-0615 + FAIL-CLAIMED-NO-FEED-WITHOUT-CHECKING — 2026-06-19
- FAIL-HYPERADAR-FORYOU-SLICE-AND-TOOLS-ONLY-FILTER — 2026-06-19
- 2026-06-18 — FAIL-DEPLOY-CROSS-PROJECT-BUILD-SLOT-WEDGE (Scotty)
- 2026-06-18 — FAIL-DEPLOY-VERCEL-SHIM-FALSE-SUCCESS (recurring, root cause now fixed)
- FAIL-HYPERADAR-API-TIMELINE-LOSSY-2026-06-15