Once code got cheap, design quality is decided by the versions you say no to — a prototyping methodology you can copy
Designing in the age of the 10x engineer ft. Andy Zhang (Google) | Config 2026 · Andy Zhang
21 min total·Actually worth watching closely: ~5 min·2 must-watch clips
- 0:12 – 2:57Listen
The premise: code is no longer scarce
The opening sets up the premise for everything that follows — agents made writing code cheap, so the most expensive step in the design process has come loose.
Design quality is the accumulation of many good decisions, and cheaper code means you can afford more of them.
Mostly laying out the argument, nothing you need to watch — fine to listen to while doing something else.▶ Jump to 0:12Speaker · Andy Zhang - 2:57 – 7:13Skim
The blank spot on the tooling map
He spreads the tools used in design across two axes, time to build and cost, and points at an awkward fact: almost nothing let you validate whether an idea was right both quickly and concretely.
Inspiration references and low-fidelity sketches stay in the imagination and can't give you concrete confidence — that blank spot is exactly where prototyping belongs.
The screen shows a two-axis chart; a glance at which quadrant each kind of tool lands in and where the gap is will do, since the argument itself follows the narration.▶ Jump to 2:57Speaker · Andy Zhang - 7:13 – 10:36Listen
Spend the speed on iteration, not an earlier launch
The time AI saves should buy more rounds of iteration and learning rather than pulling the launch date forward. Paired with a gated internal rollout, only the features that hit the bar graduate to the outside world.
The value of a prototyping process isn't shippable code — it's getting to the answers for why and why not faster.
All spoken argument about process and team mechanics; the visuals don't help, so just follow the logic.▶ Jump to 7:13Speaker · Andy Zhang - 10:36 – 12:27Watch
Starting point one: from scratch
The first place a prototype can start is a blue-sky proof of concept. The live example syncs the agent's answers into a FigJam diagram, using a diagram in place of long pages of text.
Build the prototype with a specific question in hand — like whether agents could interact with you in a visual diagram — and only then does it have a reason to exist.
There's a working thing running on screen; how the diagram grows and how it tracks the conversation is obvious to watch and easy to miss by ear.▶ Jump to 10:36Speaker · Andy Zhang - 12:27 – 15:50Skim
Starting points two and three: the playground and forking production
The second is building in a playground isolated from production, where you start with baseline quality and ready-made building blocks; the third is forking the production code outright, closest to real usage at the cost of environment and credential setup. He also mentions Brian Lovin's playground at Notion, with its one-click deploy and a sub-agent that checks work against the design.
If you don't have a prototyping playground, go build one — it determines how high your starting line is on every iteration.
A few playground screenshots flash by; a look at what capabilities come pre-loaded is enough, and which starting point suits which question is mostly spoken.▶ Jump to 12:27Speaker · Andy Zhang - 15:50 – 18:37Watch
Curation: most prototypes shouldn't reach users
He makes the case with a prototype of his own that got cut — a clickable interface where you could comment directly on the graph. It was cool, but he judged he wouldn't actually use it, so he dropped it. That leads into shipping as an act of curation.
Working and cool don't add up to should-ship; cheap code is a reason to collect more feedback, not a reason to ship more.
The rejected interface is right there on screen — you have to see that it really was well made for the "but I wouldn't use it" to carry any weight.▶ Jump to 15:50Speaker · Andy Zhang - 18:40 – 20:35Listen
Tools don't beat insight
The close comes back to judgment: quality isn't about how much you ship or how close you are to production code, but about which version you nodded at and which ones you shook your head at.
Don't let AI decide what it can't verify — it isn't your user, and without a prototyping process you have no compass for finding quality.
Pure closing argument over a few sparse text slides; listening to the last two minutes is enough.▶ Jump to 18:40Speaker · Andy Zhang