2026-05-06

• #0012

  #0012Indie businessr/SideProject2 weeks ago

  A 3-day `Wordle` spinoff became a durable **$3k/mo** ad business

  80radar

  Distribution, not polish, made this work. A built-in share loop turned a throwaway daily game into a durable cash-flow asset worth copying.

  • —Launched rough and got bad early feedback, but the first stable audience came from a niche newsletter, not the initial post.
  • —The real growth lever was a shareable score button; once people posted results, larger accounts and media amplified it.
  • —The project still does ~10k DAU and ~$3k/mo from display ads, with low ongoing maintenance.
  • —A simple consumer daily game proved viable without turning into B2B SaaS, especially when replay and sharing are built in.

  Source: www.reddit.com/r/SideProject/comments/1t5d028/4_years_agRead original →
• #0011

  #0011Agents & toolsAnthropic News2 weeks ago

  `Claude Code` doubles 5-hour limits, drops peak-hour throttling

  80radar

  Claude CodeCoding agent — longer session limits for sustained work

  Heavy users can stay in session longer and hit fewer slowdowns. Better Claude Code throughput now; the SpaceX compute story matters only because it unlocks higher real limits.

  • —Paid Claude Code tiers now get 2x the five-hour quota, which directly reduces session cutoffs during long coding runs.
  • —Pro and Max no longer face peak-hour limit reductions, so daytime usage becomes more predictable.
  • —Claude Opus API rate limits also rise considerably, improving headroom for agent loops and batch workloads.
  • —SpaceX's Colossus 1 adds 300MW+ and 220,000+ NVIDIA GPUs within a month, with capacity routed to paid subscribers first.

  Source: www.anthropic.com/news/higher-limits-spacexRead original →
• #0010

  #0010Agents & toolsClaude Code Releases2 weeks ago

  `Claude Code` `v2.1.129` adds session plugin fetch and background auto-update

  70radar

  Claude CodeAI coding tool — terminal-first dev workflow

  Plugins are easier to load on the fly, and self-updates now run in the background on package installs. Small release, but the workflow friction drop is real enough to use now.

  • —--plugin-url <url> pulls a plugin .zip into the current session, which lowers the setup cost for trying custom tooling fast.
  • —CLAUDE_CODE_PACKAGE_MANAGER_AUTO_UPDATE lets Homebrew and WinGet installs upgrade in the background, then asks for a restart instead of blocking work.
  • —Gateway model discovery is now opt-in via CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1, reducing surprise model list changes on custom gateways.
  • —skillOverrides now behaves correctly across off, user-invocable-only, and name-only, which makes shared agent setups easier to control.
  • —Several regressions were fixed around /clear, subagent UI, prompt cache TTL, and MCP allow rules, so day-to-day CLI reliability improved.

  Source: github.com/anthropics/claude-code/releases/tag/v2.1.129Read original →
• #0009

  #0009Indie businessr/SaaS2 weeks ago

  Cold-call funnel from 1,541 leads to 82 closes and $23.5k

  60radar

  Brute-force outbound still works when the funnel is tight and measured. 1,327 calls turned into 82 closes, which makes cold calling look viable if you can tolerate rejection and track conversion at each step.

  • —The funnel is unusually transparent: 1,541 leads → 1,327 calls → 613 answers → 82 closes → $23,487.
  • —Answer rate was about 46% of calls, and closes were about 13% of answers. That is strong enough to justify manual outbound before buying ads.
  • —The useful pattern is not motivation but instrumentation: track each stage, find where drop-off happens, and optimize the weak link first.
  • —This is a reminder that distribution can be learned late. A technical founder can brute-force sales motion before building a larger GTM stack.

  Source: www.reddit.com/r/SaaS/comments/1t50w9t/got_laid_off_fromRead original →
• #0008

  #0008Agents & toolsGitHub Changelog2 weeks ago

  `GitHub MCP Server` secret scanning is now generally available

  60radar

  GitHub MCP ServerMCP server — GitHub workflows with built-in secret scanning

  Secret scanning now lands directly inside MCP-based coding workflows. If you rely on agentic IDEs or CLIs, this is an immediate security upgrade with almost no adoption friction.

  • —GitHub MCP Server now supports secret scanning in GA, so the feature is no longer experimental.
  • —The target surface is MCP-compatible AI tools, including examples like GitHub Copilot CLI and agent-friendly IDE flows.
  • —This pushes security checks closer to where code is generated, which matters more as more edits originate from agents.

  Source: github.blog/changelog/2026-05-05-secret-scanning-with-giRead original →
• #0007

  #0007Idea signalsr/ClaudeAI2 weeks ago

  Gap in Word Familiarity Scoring for Games and Learning Apps

  50radar

  Frequency scores like Zipf miss whether adults actually know a word. That leaves a clear product gap for familiarity scoring in word games, edtech, and search UX.

  • —wordfreq-style frequency can underrate words people know but rarely type, such as stethoscope, campanile, and agave.
  • —Plural or inflected forms like playbills, grottoes, and crunches can look obscure in corpora even when the base word is common.
  • —The missing metric is recognition, not production frequency. A separate familiarity layer would improve hinting, difficulty curves, and validation.
  • —This points to a small data product opportunity: an API or scoring dataset tuned for human word familiarity rather than internet text frequency.

  Source: www.reddit.com/r/ClaudeAI/comments/1t5dfjn/kindergartengRead original →
• #0006

  #0006Agents & toolsr/LocalLLaMA2 weeks ago

  `llama.cpp` MTP makes `Qwen 3.6 27B` far more usable for local coding agents

  50radar

  llama.cppInference engine — lightweight local LLM serving

  A custom llama.cpp build stacks MTP, turbo4 KV cache, and 262K context on 48GB Macs. Still a manual setup, but local agentic coding just moved from hobbyist tweak to viable option.

  • —--spec-type mtp --spec-draft-n-max 5 delivered 2.5x faster generation, reaching 28 tok/s on an M2 Max 96GB.
  • —turbo4 KV cache cuts KV memory to roughly one quarter, which is the real unlock for long-context local use.
  • —A 262K context window reportedly fits on 48GB Apple Silicon with Q5_K_M plus turbo4, making repo-scale sessions more realistic.
  • —The package also ships fixed chat templates and llama-server OpenAI/Anthropic-compatible endpoints, so existing agent stacks need less glue code.

  Source: www.reddit.com/r/LocalLLaMA/comments/1t57xuu/25x_faster_Read original →
• #0005

  #0005Infra & SaaSSupabase Blog2 weeks ago

  `@supabase/server` bundles server auth, `RLS` scope, and `CORS`

  50radar

  Server auth, RLS-scoped clients, and CORS are packaged together. Clear boilerplate reduction if most of your data access lives on the server.

  • —Stateless auth is built in, which lowers session-handling glue in server-heavy flows.
  • —RLS-scoped clients on the server keep permission boundaries closer to the query path.
  • —CORS handling ships with the package, so custom middleware and repeated setup shrink.

  Source: supabase.com/blog/introducing-supabase-serverRead original →
• #0004

  #0004Infra & SaaSvercel_blog2 weeks ago

  `Vercel` adds production-only access for Marketplace credentials

  50radar

  Marketplace credentials can now be locked to Production, cutting off dashboard and CLI reads outside prod. Useful guardrail if preview or dev projects should never touch live secrets.

  • —Turning on Allowed Environments -> Production only removes non-production access and protects secret values from dashboard and CLI reads.
  • —After the switch, existing Development and Preview connections are removed, and any connection without a Production target is disconnected.
  • —New non-production connections are blocked going forward, so accidental secret exposure from test projects gets shut down at the platform layer.
  • —Rolling this back needs Owner permissions, may trigger MFA, and Vercel recommends rotating the integration secrets after enabling it.

  Source: vercel.com/changelog/secure-marketplace-credentials-withRead original →
• #0003

  #0003Agents & toolsCursor Changelog2 weeks ago

  `Cursor` adds context usage breakdown for agents

  50radar

  You can now inspect where an agent's context window is being spent instead of treating token use as a black box. Useful for trimming bloated prompts and debugging why long runs get expensive or lose focus; worth using if you lean on Cursor agents.

  • —The update exposes a breakdown of agent context usage, not just a single opaque total.
  • —Better visibility helps pinpoint prompt bloat, attached files, or history that eats the context window early.
  • —When an agent derails on long tasks, this gives a direct place to check whether context pressure is the real cause.

  Source: cursor.com/changelog/05-06-26Read original →
• #0002

  #0002Idea signalsr/SideProject2 weeks ago

  YC idea pattern: chase expensive problems, not exciting ones

  50radar

  The better filter is current dollar loss, not vague market size. If you can't name the annual cost in 60 seconds, the idea is probably still too soft.

  • —Across YC requests, the recurring formula is stable: big industry, costly pain, new technical unlock, then build there.
  • —Examples skew boring on the surface but huge in spend: payroll, credit cards, HR admin, data pipes, background checks.
  • —The suggested screening question is sharper than TAM talk: what does this problem cost this year, in dollars?
  • —2026 examples cite $35B prior-auth cost, $40-50B legal overspend, and $30-40B wasted inference compute.
  • —Useful as an idea filter for AI and SaaS alike: start from budget line items, then map recent tech shifts.

  Source: www.reddit.com/r/SideProject/comments/1t4m100/i_read_eveRead original →
• #0001

  #0001Agents & toolsClaude Code Releases2 weeks ago

  `Claude Code` `v2.1.131`, Windows VS Code activation and Mantle auth fixes

  40radar

  MantleAuth service — integrates via x-api-key headers

  Two break-fix patches landed: the VS Code extension now activates on Windows again, and Mantle auth restores the missing x-api-key header. Low-drama release, but worth updating if either path was broken for you.

  • —A hardcoded bundled SDK build path plus a createRequire polyfill bug was breaking VS Code extension activation on Windows.
  • —Mantle endpoint authentication was failing because the x-api-key header was missing; this release restores it.
  • —No new feature here, just reliability work. Update quickly if you depend on Windows editor flows or Mantle integrations.

  Source: github.com/anthropics/claude-code/releases/tag/v2.1.131Read original →