I Think We're Losing Control Of AI

Claude 5 Fable feels like a step change in AI capability, and I cannot shake the feeling that something fundamental just shifted. In this video, I walk through my experience as an unemployed ex big tech software engineer with 25 years in the industry, exploring what happened when I got access to Claude Fable, a supposedly “sanitized” consumer version of Anthropic’s more powerful frontier model known internally as Mythos. According to public narratives, Mythos is so advanced that access is restricted to government programs and select companies under initiatives like Project Glasswing, due to potential risks if it were widely released. Fable, by contrast, is supposed to be heavily safety filtered. A downgraded version. A safe glimpse into something much larger. So I decided to push it. Over the course of two days I tinkered with Fable. In one example, I gave it a single vague prompt to build a modern isochrone mapping application for the United States, dynamically adapting travel times based on location and transportation mode. No detailed spec, no step by step instructions, just a high level goal. What followed genuinely surprised me. The model independently researched transportation timing data, designed system architecture, generated production quality code, and produced a fully working application in under an hour. It even asked me nuanced business level questions about whether to include walking time to transit hubs and how to model traffic congestion, showing a level of judgment I did not expect from an open ended prompt. Under the hood, the system appeared to orchestrate multiple sub agents performing research, architecture review, code review, and QA testing in parallel. Essentially a full AI engineering team coordinated by a single model. This is not vibe coding. This feels closer to autonomous software delivery. I also discuss the implications of this shift. The cost is extremely high, the safety guardrails are restrictive in unexpected ways, and the biggest concern is the loss of human visibility into decision making. The system is doing real cognitive work, but much of it is opaque to the user. Using an orchestra analogy, we are no longer conducting every note. We are setting intent and watching an entire performance unfold from systems we only partially understand. The conclusion is uncomfortable. Models like Claude Fable are no longer simple tools. They behave like emerging cognitive systems capable of research, planning, reasoning, and execution at a level that begins to resemble human teams. Whether this becomes the most powerful productivity leap in history or something far more difficult to control is still an open question. Timestamps: 0:00 Intro 2:16 A Test of Fable's capabilitis 4:33 Some amazing things I saw 9:05 Some real challenges 9:30 The job made me waste money 11:10 The leviathan awakens? Join this channel to get access to perks:    / @asiandadenergy   Buy me a coffee: https://buymeacoffee.com/codeslingem Subscribe to my Newsletter: https://asiandadenergy.substack.com/ Book a coaching session with me: https://calendly.com/codeslinger1 Buy my merch: https://asian-dad-energy-shop.fourthw... #aitransformation #bigtech #softwareengineer #agi #claude #claudeai #fable #mythos