Anthropic's Dangerous Model + The AI-for-SMB Rush
Anthropic built a frontier model so dangerous they refused to ship it. Responsible safety call, or 2026's best marketing stunt? That question is where Oscar and I started the whole show — the very first Signal or Noise call on the very first episode of Human in the Loop.
Here's how I read the week that kicked it all off.
Signal or Noise
Claude Mythos Preview — the model "too dangerous to ship"
I'll be honest: "too dangerous to release" is also the best product teaser in the industry. Both things can be true. Watch what they do over the next few months, not what the announcement says. Signal, with a raised eyebrow.
Meta Muse Spark
Meta's push into the generative stack. Real, but early. Signal, lightly.
OpenAI "Spud" / GPT-5.5
A lot of pre-launch noise for a model that isn't in your hands yet. Noise until you can run it.
The OpenAI + Anthropic + Google anti-distillation pact
Three rivals agreeing to protect their outputs from being copied tells you exactly where they think the moat is — and how fragile it is. Signal.
Anthropic's $400M Coefficient Bio acquisition
A real bet on AI-for-science. Signal, lightly — worth tracking where the frontier labs point their money.
Ship It or Skip It
The three ideas we opened the series with:
- Bring AI to small businesses — the big one. We asked on day one whether there was a real business here. (Spoiler for anyone reading in order: by Episode 6, Anthropic shipped it themselves — which tells you the demand was real and the moat was never the model.)
- A vertical AI agent for insurance claims — narrow, painful, valuable. Ship it.
- Lovable-for-X — app-builder tooling aimed at a specific niche. Depends entirely on the niche. Conditional ship.
That's where Human in the Loop began: no hype, no doomerism, two builders making the call. Fifteen-plus episodes later, the framing still holds. Watch or listen to the one that started it all below.
