Agent Product & Capability Design
Centered on the five capability dimensions, with the five-layer technical foundation as the base. How to design and evaluate an Agent.
This module answers one question: how do you design and evaluate an Agent as a product?
The main thread is the five dimensions for assessing an Agent's capability — task-completion path, failure nodes, error recovery, transparency, and boundary behavior; the base is the five-layer technical foundation for understanding how an Agent runs (Loop / Tool / Planning / Memory / Multi-Agent). By the end you'll own a vocabulary that is decomposable, instrumentable, and KPI-ready: turning the vague "is this Agent any good?" into concrete dimensions you design and measure one by one.
Lecture order: Lecture 0 sets the base with the five-layer architecture (locate which layer a problem lives on), then the five dimensions unfold one per lecture — the first four build on each other (can it finish → where does it fail → how does it recover → is the process transparent), ending on boundary behavior (should it act at all).
For: product managers moving into or leveling up on Agents, model-strategy PMs, and engineers who want a "product evaluation" lens.
Posts
Why Understanding Agents Starts With Splitting Them Into Five Layers
An Agent's capabilities and failures all grow on five foundational layers — Loop / Tool / Planning / Memory / Multi-Agent.
A · 1 Product · Post 1Why an Agent's First Move Shouldn't Be to Start Doing
The task completion path is an Agent's ability to break down and drive a request from need to output. PMs should design it to be observable and instrumentable.
A · 2 Product · Post 2Why You Can't Blame Agent Failures in the Abstract — Break Them Down by Node
Split Agent failures into four node types — input / tool / reasoning / output — locate first, then improve.
A · 3 Product · Post 3Why an Agent With No Error-Recovery Strategy Falls Apart at the First Mistake
A 4+1 taxonomy, a 4-question diagnostic SOP, and an anti-pattern checklist that explain how an Agent survives after it goes wrong.
A · 4 Product · Post 4Why More Agent Transparency Isn't Always Better
Transparency isn't a scalar — it's a matrix of "audience × granularity × delivery path." Show it to the wrong audience, and transparency turns into negative value.
A · 5 Product · Post 5Why an Agent That Can't Respect Boundaries Becomes a Dangerous "Overconfident Assistant"
Break down 4 boundary types × 4 behavior modes into a designable, measurable boundary-behavior system.