Action Produces Information
Ship the smallest test to expose what thinking cannot reveal
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 99%
Armstrong says that when he did not know what to do, completing something produced information that further thought had not supplied. Before Coinbase, he and a friend built an application to understand Bitcoin. As soon as they shipped it, he could see that its architecture was wrong and knew what to do next. The reusable loop begins with a specific uncertainty, creates the smallest safe artifact capable of testing it, exposes that artifact to real use, and converts the result into the next decision. The first output is not expected to be correct or commercially complete. Its value is the new information it makes observable, provided the builder limits the test's downside and does not confuse activity with a meaningful experiment.
Origin
Armstrong credits the phrase to Paul Graham. He illustrates it with a pre-Coinbase Bitcoin application whose flaws became obvious only after he and a friend shipped it.
Core principles
- 01Uncertainty can persist when an idea exists only in thought
- 02A small completed action creates evidence about the next move
- 03The first version may be useful mainly because it reveals what is wrong
- 04Learning speed matters more than protecting the first architecture
How to run it
- 1
Name the unknown
State the question preventing a decision. Make it narrow enough that one artifact or action could provide evidence.
Pro tip Phrase it as what you need to learn, not what you hope is true.
- 2
Design the smallest test
Choose a prototype, conversation, or release that can expose the relevant information. Remove features unrelated to the question.
Pro tip Define in advance what observation would change your next move.
Watch out Do not release unsafe or misleading prototypes to real users.
- 3
Complete and expose it
Finish the test and place it in the environment where the unknown becomes observable. Avoid keeping it private merely to protect the first attempt.
- 4
Read the new signal
Identify what became clear only after action. Separate evidence about the idea from embarrassment about an imperfect first version.
Pro tip Write the newly learned fact in one sentence.
- 5
Choose the next move
Revise, replace, continue, or stop based on the signal. Start another small loop if the next decision remains uncertain.
Pro tip Let each test retire one important unknown.
Watch out Repeated motion without a defined learning question is not this framework.
In the wild
Armstrong and a friend built an application while trying to understand Bitcoin. Once it shipped, Armstrong immediately recognised that the architecture was wrong; that completed attempt gave him information about what to build next.
→ An imperfect artifact replaced abstract uncertainty with a concrete architectural lesson.
Common mistakes
Polishing before learning
Extra work that cannot affect the target question delays the information the test is meant to produce.
Acting without a question
Busywork produces output, but not necessarily information relevant to a decision.
Ignoring test downside
A learning objective does not excuse exposing others to avoidable harm or deception.
Is it for you?
Best for
Builders who cannot reason their way to the right product, architecture, message, or next step.
Not ideal for
High-risk tests where a rough release could create material safety, legal, privacy, or financial harm.
From the transcript
“action will produce information”
“the minute we shipped it i was like it's built wrong”
From the episode
Coinbase Founder: The Crazy Journey Of Building A $100 Billion Company: Brian Armstrong