The Rundown · episode

AI Agents, Cheap Proto‑YouTube & Opus Roasts

goosely walks through building a streaming site, cooking AI agents with real access, and bluntly judges Opus 4.5 — plus a budget proto‑YouTube cost breakdown.

Written by AI · goosely · Oct 20, 23:54 UTC · After-show report

Chapters

  1. 02:10 Making a streaming schedule
  2. 04:02 OpenVibe: risks of full API access
  3. 11:13 OpenVibe: Cloudflare + Backblaze routing
  4. 13:38 AI and vibe coding: how AI changes dev
  5. 18:06 AI agent applying to jobs
  6. 27:10 Which AI models to pay for
  7. 28:34 Opus 4.5 reaction: brutally honest
  8. 29:50 Cheap proto‑YouTube cost breakdown

What happened

  1. goosely says basic scripting knowledge is enough to start a site

    goosely argues that knowing how to compile small Python scripts into an executable and basic scripting knowledge is sufficient to begin building a streaming website or really to start many projects.

    goosely

  2. goosely breaks down risks of handing AI full API access

    goosely warns that while you could hand an AI full API tokens (e.g., Cloudflare) to do tasks, it's not recommended, though a competent model probably won't 'fuck shit up.'

    goosely

  3. goosely explains OpenVibe's storage routing setup

    goosely describes OpenVibe's more intricate streaming setup using Cloudflare and Backblaze, how popular content is served from Cloudflare while less popular content gets pulled from Backblaze, and mentions temporary hot storage on the streaming server that gets offloaded.

    goosely

  4. Goosely on AI making coding easier — and worrying

    goosely talked about how AI has made programming much easier for newcomers, saying prior coding experience now just helps you know what to ask and that basic 'code theory' suffices more than deep tooling knowledge.

    goosely

  5. goosely sets their AI loose to apply for jobs

    goosely described setting up an AI agent to apply to jobs and create accounts on hiring sites, and noted it now has some access to their email/contacts for that work.

    goosely

  6. Which AI models are worth paying for, according to goosely

    goosely compared AI models they use for projects, saying older free models are weak and recommending cheap paid models like Deep Seek 4.1, Opus 5.5, JetGPT Astra and mentioning Gemini/Argon options and promotional pricing.

    goosely

  7. Goosely reacts to Opus 4.5 image

    goosely looks at an image they think might be from Opus 4.5 and gives a running, blunt reaction to how it looks.

    goosely

  8. Why goosely's site uses WebRTC for low-delay streams

    goosely explains the site's streaming protocols: it supports WebRTC for low latency and another ingest/transcoding flow (they misname terms but describe trade-offs). They say WebRTC gives minimal delay but fewer quality options, while traditional RTMP/transcode approaches add delay but allow quality settings.

    goosely

  9. goosely on hosting costs and robot control needs

    goosely describes their two robots (Cosmo and a Raspberry Pi bot) and why low stream delay matters for user-controllable bots. They outline hosting choices, saying storage is cheap with Backblaze and that the site can serve hundreds to a thousand concurrent viewers on modest costs.

    goosely

  10. How cheap could a proto-YouTube be?

    goosely walks through rough hosting and bandwidth costs for a small, early-YouTube style site, repeatedly saying the answer depends on traffic and desired guarantees but offering concrete monthly estimates for budget setups.

    goosely

  11. goosely says Claude is the high-performance bar — but pricey

    goosely ran through their experience using Claude versus cheaper APIs, saying Claude (on the $200 plan) is the performance benchmark for high-end AI work but can be expensive and hit usage caps. They explained they mix Claude with other tools for broader projects like OpenVibe and sometimes override APIs to save costs.

    goosely

  12. goosely recommends ChatGPT Plus for heavy AI work

    goosely says the ChatGPT paid plan (Plus) is worthwhile for heavy AI users, giving access to codex usage and helpful tools for working with open-source codebases and internal wikis.

    goosely

More from goosely