2026-05-05

• #0014

  #0014Indie businessr/SideProject2 weeks ago

  Hybrid AI media network hit $2.14M pre-tax with one-operator automation

  80radar

  This is less about prompts and more about building a deterministic content machine with clear human decision points. The revenue mix is strong, but the real takeaway is that asset quality and channel reputation drive long-term resale value.

  • —Revenue split was $2.14M pre-tax: $1.22M from sponsorships/ads, $430k from platform bonuses, $490k from selling five history pages.
  • —Scale came from a mostly solo operation running a 36M-follower network and over a billion monthly views; the leverage is systems design, not team size.
  • —The stack started from internal workflow automation, then moved to owned media when corporate ROI proved weak. Build for owned upside, not borrowed salary gains.
  • —He maps each workflow node first, then automates repeatable steps and keeps taste-sensitive choices manual. AI is the generation layer, not the whole business.
  • —Short-term monetization can damage the asset. OnlyFans promotions paid well but hurt account reputation, which directly cuts long-term earnings and exit value.

  Source: www.reddit.com/r/SideProject/comments/1t4f8s2/ive_been_rRead original →
• #0013

  #0013Models & APIOpenAI2 weeks ago

  `GPT-5.5 Instant` becomes ChatGPT’s new default model

  80radar

  ChatGPT now defaults to a model tuned for better accuracy, fewer hallucinations, and tighter personalization control. That shifts day-to-day output quality without any workflow change, so it is worth re-checking prompts and saved instructions now.

  • —ChatGPT switched its default model, so baseline answer quality changes even if users do nothing. Old prompt behavior may no longer map 1:1.
  • —OpenAI highlights smarter, more accurate responses and fewer hallucinations. That directly affects drafting, coding help, and support-style flows.
  • —Personalization controls improved alongside the model update. Saved instructions and tone steering likely matter more in repeat-use workflows.

  Source: openai.com/index/gpt-5-5-instantRead original →
• #0012

  #0012Infra & SaaSvercel_blog2 weeks ago

  `Vercel CLI` adds `vercel metrics` for Observability queries

  60radar

  Observability data is now queryable from the terminal instead of only the dashboard. Useful for agent-driven debugging and ops workflows, but gated behind Observability Plus so the impact is narrower.

  • —The new vercel metrics command exposes Observability Plus metrics in the CLI, so performance and reliability checks can fit into terminal workflows.
  • —Queries work at both team and project scope on Vercel, which makes shared infra monitoring easier without bouncing through the web UI.
  • —Vercel explicitly positions this for coding agents too, opening the door to automated debugging loops tied to live production metrics.
  • —Availability is public beta and limited to teams on Observability Plus, so this is not a default win for every Vercel project yet.

  Source: vercel.com/changelog/vercel-metrics-in-cliRead original →
• #0011

  #0011Agents & toolsClaude Code Releases2 weeks ago

  `Claude Code` `v2.1.128` adds ZIP plugins and tighter MCP diagnostics

  60radar

  Claude CodeCoding CLI — smoother plugin and MCP operations

  Small but practical CLI update: ZIP plugin archives, better MCP visibility, and safer subprocess env handling land together. Worth updating if you rely on plugins, MCP servers, or long-running shell workflows.

  • —--plugin-dir now accepts `.zip` plugin archives, which removes manual unpacking and makes plugin distribution cleaner.
  • —/mcp now shows tool counts and flags servers connected with 0 tools, cutting time spent debugging broken MCP setups.
  • —Subprocesses stop inheriting OTEL_* vars, so Bash, hooks, MCP, and LSP runs no longer accidentally leak the CLI's telemetry endpoint.
  • —EnterWorktree now branches from local HEAD, preventing unpushed local commits from being dropped during worktree creation.
  • —Parallel shell calls are more robust: one failing read-only command no longer cancels sibling calls, which matters for multi-step repo inspection flows.

  Source: github.com/anthropics/claude-code/releases/tag/v2.1.128Read original →
• #0010

  #0010Agents & toolsMCP Servers2 weeks ago

  `MCP Servers` Python workspaces patch multiple security alerts

  50radar

  MCP ServersMCP server collection — official Git, Fetch, and Time tools

  Security fixes landed across git, fetch, and time workspaces. Worth updating if you run official MCP servers; no product-level feature change.

  • —gitpython moved from 3.1.45 to 3.1.49, closing high-severity command injection paths through Git options validation.
  • —lxml 5.3.0 to 6.1.0 fixes XXE exposure in parser defaults; fetch servers handling untrusted XML should update quickly.
  • —pyjwt 2.10.1 to 2.12.1 closes unknown crit header acceptance across all three Python workspaces.
  • —Verification is solid: uv run pytest passed with 20 fetch, 38 time, and 41 git tests; Windows cleanup errors were preexisting.

  Source: github.com/modelcontextprotocol/servers/commit/0db9a3407Read original →
• #0009

  #0009ASO & growthOpenAI2 weeks ago

  `ChatGPT` ads add self-serve buying and `CPC` bidding

  50radar

  Ads ManagerAd platform — self-serve buying with CPC bidding

  OpenAI is turning ChatGPT ads into a real performance channel with a beta self-serve manager, CPC bidding, and better measurement. Useful once traffic quality is proven; for now, worth watching more than rushing budget in.

  • —A beta self-serve Ads Manager removes the sales-led bottleneck, so smaller teams can test without enterprise-style ad buying.
  • —CPC bidding shifts the product toward performance marketing instead of pure sponsorship, making spend easier to control.
  • —Enhanced measurement tools matter because attribution decides whether this becomes a real acquisition channel or just branded exposure.
  • —OpenAI says conversations stay separate from ads and privacy is protected, which lowers targeting depth and may limit precision.

  Source: openai.com/index/new-ways-to-buy-chatgpt-adsRead original →
• #0008

  #0008Agents & toolsr/ClaudeAI2 weeks ago

  `Relay`: message parallel `Claude Code` sessions instead of alt-tabbing

  50radar

  RelayPlugin — local messaging across Claude Code sessions

  Local session-to-session messaging removes alt-tab copy-paste. Niche, but worth trying if you already split work across repos.

  • —Uses Claude Code channels so one session can receive another session's question between turns and answer on its next turn.
  • —Architecture stays minimal: one local hub daemon, tiny per-session MCP server, routing over a Unix socket.
  • —Supports direct asks and broadcasts like 'what is everyone working on', with replies arriving one by one as notifications.
  • —Scope is intentionally narrow: same machine only, no auth, nothing leaves the box, so setup is simple but remote/team use is out.

  Source: www.reddit.com/r/ClaudeAI/comments/1t3osat/built_a_plugiRead original →
• #0007

  #0007OtherOpenAI Python SDK2 weeks ago

  `openai-python` `v2.34.0` adds Admin API Key support and env-based headers

  50radar

  openai-pythonPython SDK — better Admin key and env header support

  Project and user admin fields expanded, and auth paths were cleaned up. Upgrade if you automate admin flows or juggle multiple credentials; otherwise this is mostly a stability release.

  • —projects gets external_key_id, and users adds email and metadata params, which reduces custom wrappers around admin provisioning flows.
  • —Per-endpoint Admin API Key support landed with related auth fixes, including no bearer fallback and preserved selected credentials during requests.
  • —OPENAI header configuration via env is now supported, useful for shared deploy setups where runtime header injection was previously manual.
  • —Type and transport fixes cover Response timestamps, prompt_cache_retention, Azure auth headers, and multipart file array field names.

  Source: github.com/openai/openai-python/releases/tag/v2.34.0Read original →
• #0006

  #0006OtherAnthropic Python SDK2 weeks ago

  `anthropic-sdk-python` `v0.98.0` adds WIF, interactive OAuth, and auth profiles

  50radar

  anthropic-sdk-pythonPython SDK — unified WIF, OAuth, auth profiles

  Auth setup got much better, and Managed Agents APIs moved forward too. Useful if you juggle multiple creds or deploy across cloud environments; otherwise this is a solid maintenance upgrade.

  • —Workload Identity Federation, interactive OAuth, and auth profiles landed together, cutting secret management friction across local and cloud setups.
  • —You can now set headers via environment variables, which simplifies per-env routing, tracing, and proxy-style customization without code edits.
  • —Managed Agents APIs were improved again, a signal that Anthropic is still actively shaping higher-level agent workflows in the SDK.
  • —Streaming now carries stop_details onto the accumulated Message, fixing a state-loss bug that could break downstream control flow.
  • —Multipart file array naming and Vertex AI async multi-region base_url bugs were fixed, reducing edge-case integration failures.

  Source: github.com/anthropics/anthropic-sdk-python/releases/tag/Read original →
• #0005

  #0005Models & APIAnthropic TypeScript SDK2 weeks ago

  `Anthropic TS SDK` `0.93.0` adds WIF, interactive OAuth, and auth profiles

  50radar

  Anthropic TypeScript SDKTypeScript SDK — adds three Anthropic auth flows

  Authentication setup just got less brittle across local dev, CLI flows, and multi-account environments. Workload Identity Federation is mainly enterprise-facing, but interactive OAuth and auth profiles make real projects easier to wire up now.

  • —The release adds three auth features at once: Workload Identity Federation, interactive OAuth, and auth profiles.
  • —Workload Identity Federation reduces long-lived secret handling, but it matters most in cloud and org-managed setups.
  • —Interactive OAuth points to smoother developer login flows for local tools and internal apps using Anthropic APIs.
  • —Auth profiles help switch credentials by environment or account, which cuts friction in side projects with staging and production splits.

  Source: github.com/anthropics/anthropic-sdk-typescript/releases/Read original →
• #0004

  #0004Agents & toolsAnthropic Engineering2 weeks ago

  `Claude Code` quality regressions traced to 3 product-layer changes

  50radar

  Claude CodeCoding agent — brings Claude into terminal workflows

  The breakage came from defaults and prompts, not the model API: lower reasoning, broken session memory cleanup, and an anti-verbosity prompt. Fixed by v2.1.116; treat client defaults and prompt tuning as production-critical.

  • —On March 4, default reasoning dropped from high to medium to cut latency; Anthropic reverted it on April 7 after users preferred better output over speed.
  • —A March 26 idle-session cleanup bug kept wiping older thinking every turn after 1 hour idle, making responses forgetful and repetitive until the April 10 fix.
  • —An April 16 system-prompt change to reduce verbosity hurt coding quality and was rolled back on April 20; prompt edits alone were enough to degrade code output.
  • —The API and inference layer were explicitly unaffected. The failures sat in Claude Code, the Agent SDK, and Claude Cowork, which matters if you benchmark via product UX rather than raw API.
  • —Usage limits were reset for subscribers on April 23. Practical takeaway: pin versions, watch default-effort changes, and regression-test prompt/session behavior around coding flows.

  Source: www.anthropic.com/engineering/april-23-postmortemRead original →
• #0003

  #0003Indie businessstartups_for_the_rest_of_us2 weeks ago

  `Startups for the Rest of Us` Ep. 831: Customer Zero Pricing, Low-Awareness Markets, and SaaS Metrics

  40radar

  A tight operator-focused Q&A on early pricing, ad copy, seasonal SaaS metrics, and selling where buyers lack problem awareness. Broad, not novel, but immediately usable if you're refining positioning or monetization.

  • —It tackles whether customer zero should pay or get free access, forcing a clear tradeoff between validation speed and pricing discipline.
  • —The ad-copy section separates written from spoken copy, useful if you're testing podcast, video, or founder-led verbal channels.
  • —Seasonal transaction-fee SaaS metrics get dedicated attention, including the limits of tracking only GMV once pricing adds subscriptions plus fees.
  • —The low-awareness market segment is the sharpest part: if buyers don't feel the problem yet, education cost rises and bootstrapped sales get harder.
  • —Freelancer guidance is framed as leverage vs. drag, which matters when outside help can either speed execution or add management overhead.

  Source: www.startupsfortherestofus.com/episodes/episode-831-writRead original →
• #0002

  #0002ASO & growthbuffer_resources2 weeks ago

  Free Workflow for Scheduling `TikTok` Posts

  40radar

  Scheduling turns short-form marketing into a weekly batch task. Useful when launch clips, feature teasers, and app updates need consistent posting without paid tooling.

  • —Desktop and mobile scheduling are both covered, so the workflow fits quick edits as well as planned content batches.
  • —The direct value is time saved: hours per week can move from manual posting to product, support, or experiments.
  • —TikTok remains a top discovery channel for visual products; a free scheduler lowers the cost of testing short-form growth.

  Source: buffer.com/resources/how-to-schedule-tiktok-posts/Read original →
• #0001

  #0001Agents & toolsr/ClaudeAI2 weeks ago

  Add one hard-line sentence to stop `Claude` from overusing em dashes

  40radar

  A single verbatim rule in Claude.ai preferences or Claude.md cuts em-dash usage by roughly 98%. Cheap prompt hygiene, worth adding if you want cleaner house style.

  • —Use this exact line on its own: CLAUDE NEVER USES EM DASHES. Instead, ALWAYS use commas or hyphens.
  • —Placement matters: add it as a new standalone sentence in profile preferences or Claude.md, not buried inside other instructions.
  • —The trick is giving a replacement behavior, not just banning one pattern. commas and hyphens act as the fallback style.
  • —The reported gain is about 98%, not perfect suppression. Old-literature or academic-style generations can still leak em dashes.

  Source: www.reddit.com/r/ClaudeAI/comments/1t3nq2o/dont_like_em_Read original →