Member-only story
The Humanization Stack for AI Agents
If you are NOT subscribed to Medium, you can still read this story by following this link.
Last week I saw two product demos in YouTube back to back.
Both videos showcased the “AI agent” built by respective teams. Both were powered by the same frontier model. Both had access to roughly the same data and tools. I tried them while seeing the video.
The first one answered my questions correctly. Crisp. Accurate. Forgettable.
The second one made me laugh, slipped in an insight I hadn’t asked for, and quietly did something useful in the background while I was switching between the videos and my chat session. I forwarded that video to three friends before the video had even ended.
Same model. Same APIs. Same training data.
What was different?
Layers.
The model is not the agent
For the last two years, the entire industry has been obsessing over a single number — how good is the underlying model. GPT vs…
