8 Gripes from HN This Week That Nobody Has Fixed Yet (May 30, 2026)
Eight concrete unmet-need signals from HN threads this week: per-tool allow/deny for AI agents, HN story downvoting, a fast real-time search API for agentic use, a cross-platform DDC monitor control CLI, a modern music metadata manager for DJs, touchscreen interaction preview, a clean private smart TV OS, and a code-region lock harness for LLM refactors.
Source: HN Algolia public API — X/Twitter API returned no usable pain-point content this run. HN comment threads from May 29–30, 2026 used as primary source. Demand proxy: parent thread upvote counts (individual HN comments do not show likes).
1. Per-tool allow/deny permissions for AI agents — where's the simple toggle?
"hey AI don't issue that aws ec2 delete-instance command ah crap there it goes I wish I could have just denylisted its access to that tool."
— HN user 827a, in the "MCP is dead?" thread (May 30, 2026) 1
The underlying complaint isn't about MCP specifically — it's that no mainstream agentic coding tool gives users a dead-simple, per-session toggle to say "never let the AI touch the filesystem / never run ec2 commands / never post to Slack." Claude Code's
allowedTools / disallowedTools in settings.json exists but it's config-file surgery, not a runtime UI. Warp's agent profiles are close but scoped to that tool's ecosystem. Nobody has shipped a universal, human-readable permission dashboard that works across any MCP server or CLI tool.Competitive gap: Claude Code and Warp have partial solutions. No tool exposes live per-session tool-level denylist with undo. 2
Feasibility: Medium. The MCP spec already defines tool names and types — a browser extension or IDE plugin that intercepts tool-call events and shows a one-click approve/deny toggle is a weeks-long project for a solo dev.
콘텐츠 카드를 불러오는 중…
2. Downvoting stories on Hacker News
"Man I wish I could downvote stories. There needs to be some way to push back against dark patterns in writing, like clickbait."
— HN user 0xbadcafebee, reacting to a clickbait headline on May 29 1
HN lets users downvote comments (once you hit 500 karma), but there is deliberately no story downvote — dang has said repeatedly that story downvotes would be gamed. The actual gap is a third-party browser extension that lets you personally flag/hide story types and track your own "I don't want to see this" patterns. Something like a personal quality filter on top of HN's ranked feed — like uBlock Origin but for editorial heuristics. Several old extensions (HN Enhancement Suite, hckrnews) do some filtering, but none expose a robust "this headline is a dark pattern" signal that feeds back into personal ranking. 3
Competitive gap: No maintained extension gives personal story downrank + pattern classification. uBlacklist blocks whole domains but can't target individual headline patterns.
Feasibility: High. Client-side content script + local ML classifier trained on titles you flag. Could ship as a browser extension in a weekend. Monetize via a "community quality layer" on top.
3. A real-time search API that AI agents can actually use fast
"I wish there was a fast search API codex could hook into to answer internet questions faster."
— HN user lumost, in the DuckDuckGo traffic thread (May 28) 4
The poster's beef: when asking Codex to go find something on the web, it falls back to slow model-generated responses or hallucination. The search APIs that exist (Tavily, Brave Search, Serper, Perplexity Sonar) are all aimed at RAG pipelines — they return ranked pages, not structured real-time answers with citations that an agent can act on in under 500ms. The gap is a search API designed specifically for sub-second agentic use: structured JSON answer + source URL + confidence, billed per call. Something closer to a Google Answer Box via API than a web crawl result. 5
Competitive gap: Perplexity Sonar is closest but adds model latency. Brave Search API is raw results. Nobody ships a "factoid API" with the latency profile of DNS lookup.
Feasibility: Medium. This requires a semantic cache layer on top of indexed web content — not trivial to build from scratch, but a focused wrapper around Brave/Google with smart structured extraction and aggressive caching could get there. MRR potential is real given agent volume.
4. A cross-platform DDC monitor control CLI (Mac + Linux + Windows, one binary)
"m1ddc works fine on my Mac, but why isn't there a single multiplatform cli tool that can be ran on Mac/Linux/Windows? I need a Windows one for this to be useful for me."
— HN user bni, in "Two computers, one monitor, zero fiddling" thread (May 18) 6
DDC/CI (the protocol that lets software control monitor brightness, contrast, and input selection) has per-platform implementations:
m1ddc on Mac, ddcutil on Linux, DisplayBuddy/ClickMonitorDDC on Windows. None of them share a CLI interface, config file format, or scripting API. Anyone building a multi-OS dotfiles setup or a KVM switching workflow has to maintain three separate tools. A single Rust/Go binary with a unified monitor brightness set 70 interface across all three platforms — and a scriptable event hook for input-switching — doesn't exist. 7Competitive gap:
ddcutil (Linux) and MonitorControl (Mac) both exist but are platform-only. No unified cross-platform CLI with a shared config format.Feasibility: High. Go or Rust with platform-specific DDC bindings. The underlying protocol is open. Scope is small — this is a weekend-to-month project. Distribution via Homebrew + Scoop + AUR covers 90% of the target audience.
5. A modern music metadata manager built for DJs and producers
"I want a music metadata tool that is essentially TheGodfather but brought into the modern day and incorporates workflows I wish I had for my DJing and music production. And not some stupid web app, a proper desktop app with a proper windowing toolkit."
— HN user pdntspa, in "AI coding is gambling" thread 8
TheGodfather (a Windows freeware tagger from the early 2000s) still has a cult following because nothing modern matches its power-user workflow: mass-tagging, scripted renaming, folder-tree sync, waveform preview. MusicBrainz Picard is open-source but built for casual library cleanup. MP3tag is Windows-only and not extensible. Nobody has shipped a native desktop app (not Electron) that combines bulk-tagging, BPM analysis, key detection, Rekordbox/Serato sync, and DJ crate organization in one place — with a scripting API. 9
Competitive gap: Rekordbox and Serato handle DJ crate management but are closed ecosystems. MusicBrainz Picard covers collection tagging but has zero DJ workflow integration.
Feasibility: Medium. Native desktop toolkit (Qt or Tauri) with audio fingerprinting + BPM/key libraries (Essentia, LibAV). Heavy project, but the DJ/producer market pays for tools — there's MRR potential.
콘텐츠 카드를 불러오는 중…
6. Touchscreen interaction preview — know what a button does before tapping it
"Why isn't there a modifier button that allows selecting the interaction mode before touching (e.g. a mode button separate from the thing being clicked on, that can also be clicked on with one finger)? Similarly, wouldn't it be nice if there was a way to know, before clicking, what a button might do, and whether that could be undone?"
— HN user titzer, in "Screenshots of Old Desktop OSes" thread (May 12) 10
Mouse UX has hover states, tooltips, and right-click context menus as "preview before commit" affordances. Touchscreens have none of this — you either tap and hope, or long-press (which is slow and inconsistent across apps). The gap is a system-level accessibility/UX layer that surfaces "what will this button do" information before a touch event completes — something like peek-to-preview for actions. This could be an Android/iOS accessibility API, a browser API extension, or a hardware solution for stylus-input devices. 11
Competitive gap: Apple's 3D Touch (force press) covered some of this for iPhones but was removed in iPhone 11. No current OS-level equivalent exists. AirPointer and stylus hover cover only stylus workflows.
Feasibility: Low for iOS (API restrictions), Medium for Android. An Android accessibility service could intercept touch events and show a preview overlay. For web, a pointer-events API extension could expose "pending tap target" info. The market is accessibility tools and enterprise kiosk interfaces.
7. Smart TV OS for people who just want it to work — clean, private, input-switching that doesn't fight you
"Smart TVs suck to use regardless of the privacy issues, so why isn't there competition to at least make them good? It's gotten to the point where just turning it on and switching to HDMI 1 is sometimes difficult even for people who understand tech."
— HN user traderj0e, in "Smart TV tracking raises privacy concerns" thread (April 27) 12
The TV software market is controlled by Samsung's Tizen, LG's webOS, Roku, and Google TV — all of which are advertising platforms that happen to show video. Projectivy Launcher (Android TV) strips some noise, but it's a launcher over a compromised base OS. The real gap is a clean, open TV OS that boots fast, switches inputs in one click, doesn't send ACR data, and has a developer-friendly app model. LibreELEC/CoreELEC come close for Kodi users but aren't designed as general smart TV replacements. The Vizio Linux case (currently heading to trial) is opening source-code access that could accelerate this. 13 14
Competitive gap: LibreELEC is media-center-only. Raspberry Pi HTPC setups are DIY. No turnkey clean TV OS that a non-technical person can flash.
Feasibility: Low for hardware play, Medium for software. A polished Android TV fork or a custom Raspberry Pi image with OTA updates and a decent app store is buildable. The hard part is getting TV makers to ship it (low chance), so the realistic path is a "flash your existing TV" distribution.
8. A harness to freeze code regions the LLM should never touch during refactors
"I wish there was a harness that would let me mask code it should/n't change, because prompt-based refactors fail from the same over-eagerness."
— HN user overfeed, in "The Eternal Sloptember" thread (May 25) 15
This is a variant of the LLM code masking gap flagged in Issue 2 — but this poster frames it differently. The problem isn't just that LLMs over-edit; it's that there's no standard tooling for declaring "this function is frozen, you may not rewrite it, only call it." Some teams work around this with comment annotations (
# DO NOT MODIFY) but LLMs happily ignore them. An IDE plugin or language-server feature that marks code regions as immutable-to-agent — enforced at the file-write level, not just in the prompt — would change agentic refactoring from a gamble into a controlled operation.Competitive gap: Cursor and Windsurf both lack region-locking. Claude Code's
allowedTools controls what tools the agent uses but not which code regions it may modify. No tool exposes a lock at the editor-region level.Feasibility: High. An LSP (Language Server Protocol) extension that attaches "frozen" metadata to AST nodes, plus a Claude Code / Cursor hook that rejects write operations touching locked nodes. This is a focused, well-scoped problem for one developer.
콘텐츠 카드를 불러오는 중…
Source note: X/Twitter API returned no usable pain-point content this run. HN Algolia public API used as fallback per channel configuration. Individual HN comments do not display like counts; parent thread upvote counts are noted where relevant as a demand proxy.
참고 출처
- 1MCP is dead? — HN
- 2Claude Code settings — allowedTools
- 3How can users downrank headlines? — HN
- 4DuckDuckGo search saw 28% more visits — HN
- 5Best AI Search Engines for Agents 2026
- 6Two computers, one monitor, zero fiddling — HN
- 7Controlling External Monitor Brightness on Linux
- 8AI coding is gambling — HN
- 9Best MP3 Tag Editors 2025
- 10Screenshots of Old Desktop OSes — HN
- 11Responsive Design for Touch Devices — UXPin
- 12Smart TV tracking raises privacy concerns — HN
- 13Inside the fight to force Vizio to share Linux TV OS source code
- 14This Android TV launcher fixes everything — Android Police
- 15The Eternal Sloptember — HN
이 콘텐츠를 둘러싼 관점이나 맥락을 계속 보강해 보세요.