<- all tokdocs
2026.09.24@truthinbeta

The video ranks Hono and Effect as its two most niche picks; they are the two highest-volume npm packages in the entire stack

Coding
Watch on TikTok

View on TikTok ->

An 18.436009 s vertical clip at 1080x1920, 30 fps, delivered as an MP4 (mov,mp4,m4a,3gp,3g2,mj2) carrying an HEVC Main-profile video stream in yuv420p at 375,840 bps across 553 frames and an AAC HE-AACv2 stereo audio stream at 44,100 Hz and 64,645 bps across 401 frames, 447,829 bps overall, 1,032,024 bytes on disk, pulled as format bytevc1_1080p_447805-1; posted at epoch 1790208153, which is 2026-09-24 00:02:33 UTC, by @truthinbeta (uploader_id 7687461535397331981, nickname "elijah"), credited to "original sound" by "elijah", with a 110-character description that is the title plus five hashtags; at capture (yt-dlp extraction epoch 1792408371, 2026-10-19 11:12:51 UTC) the counters read 7,206 views, 305 likes, 7 comments, 28 reposts and 210 saves. I read all nine extracted frames (frame_001 through frame_009, sampled at 2 s intervals from 0 s to 16 s) and, because nine samples missed three of the ten product cards, I re-extracted the clip at 1 fps and read four more frames at 5 s, 9 s, 10 s and 13 s. The format is a talking-head shot held for the entire runtime: a young man in a black henley and dark jeans sits in a mesh office chair in a bedroom, a bed with olive-green bedding behind him, a round globe lamp on a wooden nightstand to either side, a patterned rug on the floor, holding a small handheld microphone to his mouth. Nothing else changes. The only moving content is a logo card that swaps in the upper third plus a one-word caption that tracks his speech. frame_001 at 0 s shows the opening card, purple block type reading "10/10 vibe coding stack" with a white handwritten subtitle "niche edition". The product cards then run in this order: Cursor (frame_002, 2 s, white cube logo and "CURSOR" wordmark, caption "Cursor"); Vercel (frame_003, 4 s, black triangle and "Vercel", caption "deployment,"); supabase (1 fps frame at 5 s, green lightning bolt and lowercase "supabase", caption "Supabase"); shadcn/ui (frame_004, 6 s, black rounded card with a double-slash mark and "shadcn/ui", caption "shadcn"); PostHog (frame_005 at 8 s and the 1 fps frame at 9 s, white card, blue-orange-yellow hedgehog mark and "PostHog", caption "analytics,"); Sentry (1 fps frame at 10 s, black chevron mark and "Sentry", caption "tracking,"); Bun (frame_006, 10 s, white card with a smiling steamed-bun face and "Bun", caption "Bun"); Hono (frame_007, 12 s, white card with an orange flame and "Hono", caption "layer,"); Effect (1 fps frame at 13 s, white card with a stacked-layers mark and "Effect", caption "Effect"); NixOS (frame_008, 14 s, white card with the blue lambda-snowflake and the wordmark "NixOS", caption "Nix"); and finally frame_009 at 16 s, which drops the wordmark card and shows only the dark rounded Cursor app icon with the caption "Cursor". No terminal, editor, browser, dashboard, benchmark chart, pricing table or line of code appears in any frame I read, and nothing in the clip shows any of the ten tools actually running. The spoken argument is a list delivered in one breath: here is a vibe coder's tech stack, ordered so each pick is more obscure than the last, and after naming all ten the speaker undercuts the bit by telling you to use the first and third items like everyone else does. The transcript ends cleanly: the last Whisper segment closes at 18.400 s against an 18.436009 s audio stream, so the 63 words across 4 segments all sit inside real audio and nothing was hallucinated past the end, though the ASR badly mangles five of the ten product names, which the frames correct.

The "increasingly more niche" ordering inverts at the end: Hono and Effect are the highest-volume packages named

At 0.00 s the video sets its entire structure: "a vibe coders tech stack but it gets increasingly more niche." That is a falsifiable ordering claim, not a joke. It asserts a monotonic decline in obscurity across ten picks, in the order Cursor, Vercel, Supabase, shadcn/ui, PostHog, Sentry, Bun, Hono, Effect, Nix.

The npm registry publishes download counts through its public downloads API. For the week of 2026-10-06 through 2026-10-12, the packages behind the video's picks recorded these weekly download totals: hono 74,030,176; effect 64,160,971; @sentry/node 40,038,442; @sentry/browser 37,620,225; @supabase/supabase-js 31,109,104; posthog-js 16,698,952; shadcn 11,912,750; bun 3,990,852.

Hono sits at position eight in the video's ranking and Effect at position nine, which is to say they are presented as the second-most and third-most obscure things a person could reach for. They are the two largest numbers in the set by a wide margin. Hono alone pulls more than four times the weekly downloads of the shadcn CLI, which the video places four slots earlier as a mainstream pick. The ordering does not decay, it spikes at the end.

Two caveats belong on those figures. Weekly npm downloads count installs, including transitive installs pulled in by other packages and repeated CI runs, so they measure package reach rather than how many humans chose a tool on purpose. And the packages are not strictly comparable to the products: shadcn is a CLI normally invoked through npx rather than installed as a dependency, posthog-js is only the browser SDK and excludes the server libraries, and the bun npm package is an install shim rather than the runtime itself, which is why its 3,990,852 is the lowest number in the list despite Bun being one of the better-known projects here. Nix is not distributed through npm at all and has no comparable figure.

GitHub stars, fetched from the GitHub REST API on 2026-10-19, tell a less extreme version of the same story: shadcn-ui/ui 125,970; supabase/supabase 111,777; oven-sh/bun 96,306; getsentry/sentry 45,604; PostHog/posthog 40,350; honojs/hono 32,599; NixOS/nix 17,893; Effect-TS/effect 17,334. Stars decline roughly in the video's order, with two breaks. shadcn/ui at position four outstars Supabase at position three, and Bun at position seven outstars both Sentry at six and PostHog at five by more than double. The premise holds loosely on stars and collapses on downloads.

Hono is a web framework and Effect is a runtime system; the video narrows both to one feature

At 9.64 s the video says "mono for the API layer," which the card at 13 s resolves to Hono, and at 9.64 s through 14.06 s it says "effect for typed error handling."

Hono's own documentation opens by describing it as "a small, simple, and ultrafast web framework built on Web Standards." It does not describe itself as an API layer. The docs list support across Cloudflare Workers, Fastly Compute, Deno, Bun, Vercel, Netlify, AWS Lambda, Lambda@Edge and Node.js, which is the actual pitch: one framework whose routing and middleware run unchanged on any Web-Standards runtime. The site also notes that "Hono" means flame in Japanese, which is what the orange drop in the card at 12 s represents. Calling it an API layer is not wrong so much as it describes one thing you can build with it.

Effect is further from its label. Its site positions itself as "Production Grade TypeScript" and "Reliable TypeScript for the AI era," and typed errors are one bullet among several. The error-handling claim the video makes is real and the site states it directly: "Typed errors in function signatures. Short-circuit or collect failures. Automatic retry with backoff," framed against the problem that "No more catch (e: unknown), errors are fully typed." But the same page lists four other capability areas: type-safe dependency injection with automatic resolution, fiber-based structured concurrency with parallel execution limits and automatic resource cleanup, cron-like scheduling with exponential backoff and jitter, built-in observability through OpenTelemetry tracing plus structured logging and metrics, and a unified schema layer covering runtime validation, JSON serialization and API contract generation. A developer who adopts Effect for typed errors adopts all of that, which is why it is a larger commitment than the one-line pitch suggests.

Bun is four tools, shadcn/ui is explicitly not a component library, and the card says NixOS while the claim is about Nix

Three of the ten labels name the wrong scope, and in each case the vendor's own documentation says so.

At 9.64 s the video says "one for your runtime," which the card at 10 s resolves to Bun. Bun's homepage calls itself "a fast JavaScript runtime & toolkit. All in one," and immediately specifies "Runtime, package manager, test runner and bundler in a single binary." The homepage showed Bun v1.4.3 as the current release when I fetched it in October 2026. Choosing Bun for the runtime alone leaves three of its four pieces unused, and most teams who adopt Bun adopt it for the package manager first.

At 4.64 s the video says "chat stand for UI components," which the card at 6 s resolves to shadcn/ui. The shadcn/ui documentation leads with a flat denial of the framing: "This is not a component library. It is how you build your component library." The docs describe it as "a set of beautifully-designed, accessible components and a code distribution platform," and the distribution model is the point. You do not install a package and import from it. The CLI writes the component source into your repository, where you own and edit it. The video's phrasing puts shadcn/ui in the same category as a dependency you add to package.json, which is the exact comparison its maintainers wrote the first line of their docs to refuse.

At 14.06 s the video says "and nix for reproducible builds," and the card on screen at 14 s reads "NixOS," not "Nix." The nixos.org homepage separates the two under distinct headings. Nix is "The Package Manager," described as "a tool that takes a unique approach to package management and system configuration" that "runs on Linux and macOS, next to the package manager you already have." NixOS is "The Linux Distribution," described as "a Linux distribution built around the Nix package manager and Nixpkgs." The spoken claim is correct and the site backs it directly: "Nix builds packages in isolation from each other. This ensures that they are reproducible and don't have undeclared dependencies, so if a package works on one machine, it will also work on another." The card is the error. A viewer who reads the logo rather than hearing the audio goes looking for a Linux distribution when the reproducible-build property the video is selling comes from the package manager, which installs alongside macOS and existing Linux systems without replacing the operating system.

The stack buys error tracking twice, because PostHog ships it

At 4.64 s the video assigns "post hoc for analytics" and at 9.64 s "sentry for error tracking," placing them in adjacent slots as if they cover separate ground.

PostHog publishes a dedicated error tracking product at posthog.com/error-tracking, headlined "Catch and fix issues faster with PostHog." The overlap runs the other way too. Sentry's documentation describes the platform as one that "helps developers find and fix what's broken. It captures errors, logs, traces, replays, profiles, and metrics from your application," and the product pages cover error monitoring that "automatically captures unhandled exceptions and groups similar errors into issues," distributed tracing that follows "a request from your frontend through your backend and across services," session replay offering "video-like reproductions of user sessions on web and mobile," structured logs that "live alongside your errors and traces, not in a separate tool," continuous profiling, and uptime monitoring for endpoints.

Session replay is in both products. Error capture is in both products. A stack that runs PostHog and Sentry side by side is paying two vendors for a meaningful slice of the same telemetry and instrumenting the client twice to do it. That is a defensible choice if you want Sentry's stack-trace grouping and PostHog's funnels specifically, but the video presents the pair as complementary layers when the vendors are direct competitors on part of the surface.

Vercel adds a third overlap the video does not flag. The Vercel homepage I fetched in October 2026 no longer leads with deployment at all. It opens on "Agentic Infrastructure," pitched as infrastructure "for coding agents to ship apps and agents automated by agents," with named surfaces for backends, agents and Sandbox. Deployment is still what the product does, so the video's label at 4.64 s is not wrong, but it describes the Vercel of an earlier positioning cycle rather than the one on the homepage the week the clip was posted.

"Like everyone else" overstates Cursor: 17.9 percent of the 2025 Stack Overflow sample, against 75.9 percent for VS Code

The punchline at 14.06 s is "Or just use cursor and Supabase like everyone else," and frame_009 at 16 s backs it with the Cursor app icon alone. The claim is that Cursor is the default.

The 2025 Stack Overflow Developer Survey reports IDE usage from 26,143 respondents to that section, 53.3 percent of total participants. Visual Studio Code leads at 75.9 percent. Visual Studio follows at 29 percent, Notepad++ at 27.4 percent, IntelliJ IDEA at 27.1 percent and Vim at 24.3 percent. Cursor ranks sixth at 17.9 percent.

Roughly one developer in six in that sample reported using Cursor, and more than four in five reported VS Code. "Everyone else" is using VS Code. The joke still lands as a joke, because the contrast the video is drawing is between a ten-tool stack and a two-tool stack rather than between Cursor and some other editor, but the specific tool named as the mainstream default is not the mainstream default by the largest public developer survey available.

The framing device itself has a datable origin the video does not mention. The term "vibe coding" comes from a post by Andrej Karpathy on 2 February 2025: "There's a new kind of coding I call 'vibe coding', where you fully give in to the vibes, embrace exponentials, and forget that the code even exists." Collins English Dictionary named it Word of the Year for 2025 on 6 November 2025, and Merriam-Webster had listed it as a slang and trending expression in March 2025. I reached those dates through the Wikipedia entry on vibe coding rather than through Karpathy's original post or the Collins announcement, so treat the exact wording of the quote as one step removed from source.

The Whisper transcript gets five of the ten product names wrong; the frames are the authoritative record

Anyone working from transcript.txt alone would mis-source half this video. The ASR output renders Vercel as "Brucell" at 0.00 s, shadcn as "chat stand" at 4.64 s, PostHog as "post hoc" at 4.64 s, Bun as "one" at 9.64 s, and Hono as "mono" at 9.64 s. Cursor, Supabase, Sentry, Effect and Nix survive intact, though all five arrive lowercased.

Two of those errors are the dangerous kind, because the wrong reading is a plausible English phrase. "Post hoc for analytics" parses as a statistics term rather than a vendor name, and "one for your runtime" parses as a numeral. Neither trips a reader the way "Brucell" does. The logo cards resolve all five without ambiguity, which is the reason the 2 s frame sampling was not sufficient here: at nine frames it missed the Supabase card at 5 s, the Sentry card at 10 s and the Effect card at 13 s entirely, and three of the ten products would have had no visual confirmation at all.

The timing check is clean in the other direction. The final transcript segment spans 14.060 s to 18.400 s, and ffprobe reports the audio stream at 18.436009 s and the video stream at 18.433333 s. The transcript stops 36 milliseconds inside real audio, so there is no invented tail here.

Key Takeaways

  • Correction: The video's organizing claim at 0.00 s is that the stack "gets increasingly more niche," but hono (74,030,176) and effect (64,160,971) weekly npm downloads for 2026-10-06 through 2026-10-12 are the two largest figures among all the packages it names, larger than @sentry/node (40,038,442), @supabase/supabase-js (31,109,104) and shadcn (11,912,750), despite sitting in the eighth and ninth of ten slots.
  • Correction: The card at 14 s displays the NixOS wordmark while the audio at 14.06 s credits "nix for reproducible builds." nixos.org separates them: Nix is "The Package Manager" that "runs on Linux and macOS, next to the package manager you already have," and NixOS is "The Linux Distribution." The reproducible-build property belongs to Nix and does not require adopting the distribution.
  • Correction: The video calls shadcn/ui a source of "UI components" at 4.64 s. Its documentation opens with "This is not a component library. It is how you build your component library," and describes a code distribution platform that writes component source into your repository rather than a dependency you install.
  • Partial correction: "one for your runtime" at 9.64 s resolves to Bun, which calls itself "a fast JavaScript runtime & toolkit. All in one" and specifies "Runtime, package manager, test runner and bundler in a single binary." Runtime is one of four pieces in the same binary, and v1.4.3 was current in October 2026.
  • Partial correction: "mono for the API layer" resolves to Hono, which its own docs call "a small, simple, and ultrafast web framework built on Web Standards," portable across Cloudflare Workers, Fastly Compute, Deno, Bun, Vercel, Netlify, AWS Lambda, Lambda@Edge and Node.js.
  • Partial correction: "effect for typed error handling" is accurate as far as it goes. Effect's site does promise "Typed errors in function signatures" and no more catch (e: unknown), but the same page lists dependency injection, fiber-based structured concurrency, scheduling with backoff and jitter, OpenTelemetry observability and a unified schema layer. Adopting it for error types means adopting all of it.
  • Correction: "Or just use cursor and Supabase like everyone else" at 14.06 s names Cursor as the default. In the 2025 Stack Overflow Developer Survey's IDE section (26,143 respondents, 53.3 percent of participants), Cursor ranks sixth at 17.9 percent behind Visual Studio Code at 75.9 percent, Visual Studio at 29 percent, Notepad++ at 27.4 percent, IntelliJ IDEA at 27.1 percent and Vim at 24.3 percent.
  • Clarification: PostHog and Sentry are presented as separate layers at 4.64 s and 9.64 s, but PostHog ships its own error tracking product and Sentry ships session replay, logs, tracing, profiling and uptime alongside error monitoring. Running both means paying twice for an overlapping slice of telemetry.
  • Clarification: Vercel's homepage in October 2026 leads with "Agentic Infrastructure" for "coding agents to ship apps and agents automated by agents," naming backends, agents and Sandbox, rather than leading with deployment as the video's 4.64 s label implies.
  • Verified: Supabase does describe itself as a backend, so the 5 s card and the 4.64 s label match the source. supabase.com calls itself "the Postgres development platform" and "an open source backend," tagged "Build in a weekend. Scale to millions.," with Database, Auth, Storage, Edge Functions, Realtime and Vector as named products.
  • Verified: The Whisper transcript's last segment ends at 18.400 s against an 18.436009 s audio stream and an 18.433333 s video stream, so no segment is timestamped past the end of real audio and nothing in the transcript is invented tail.
  • Context: transcript.txt mis-transcribes five of the ten product names (Vercel as "Brucell," shadcn as "chat stand," PostHog as "post hoc," Bun as "one," Hono as "mono"). The on-screen logo cards are the authoritative record, and the default 2 s frame sampling missed the Supabase, Sentry and Effect cards entirely, which is why a 1 fps re-extraction was needed.
  • Context: The term the video is built on was coined by Andrej Karpathy on 2 February 2025 and named Collins English Dictionary's Word of the Year on 6 November 2025. I confirmed both dates through Wikipedia rather than the original post or the Collins announcement.
  • Unverified: Nothing in any of the thirteen frames I read shows a terminal, editor, dashboard, benchmark or running application, so the video offers no evidence that any of the ten tools were used together, and no claim about how the stack performs in combination can be checked against it.

Resources

Published September 24, 2026. Writeup generated from a favorited TikTok.