The case for showing the work

Demos over memos.

A memo describes a thing that might work.
A demo is a thing that works.

Try it

Which one tells you if it feels right?

The memo
Spec · Settings toggle

The toggle should feel responsive and satisfying. The knob slides across in 150 ms with an ease-out curve, and the track turns green to confirm.

The demo Live

Tap the switch. Drag the speed. Tap again.

Why it works

People remember what they see.

10,000

images, each shown once.
Over 80% recognized later.

Standing, 1973
+ >

Words with pictures beat
words alone, in repeated experiments.

Mayer, multimedia principle

December 9, 1968

The Mother of All Demos.

A memo describes the landscape on paper.
A running build is ground you can stand on.

90 minutes, live | Doug Engelbart Institute

Boehm’s spiral model, 1988

Every turn ends in a prototype.

Each turn costs more than the last.
The prototype earns the next one.

Boehm's spiral model, simplified: four quadrants, a spiral widening outward over four turns, with a prototype on every turn. 1 Objectivesgoals and constraints 2 Riska prototype every turn 3 Build and testthe next version 4 Plancommit to the next turn 1234
Prototype Distance from center = cumulative cost
Boehm, IEEE Computer, 1988 | Redrawn and simplified

John Boyd’s OODA loop

Only one loop touches reality.

A demo is an act.
Every act brings back something new to observe.

Demo

The demo loop: observe, orient, decide, act, and every act produces new evidence to observe. newevidence ObserveOrientDecideAct closes on reality
Each cycle ends in contact with the product.

Memo

The memo loop: it cycles between orienting and deciding, redrafting, and never reaches act or observe. neverreached redraft ObserveOrientDecideAct closes on itself
Each draft ends in another draft.

Boyd’s loop is a model from military strategy, not a controlled study.

Boyd, The Essence of Winning and Losing, 1996 | Osinga, Science, Strategy and War

Nielsen & Landauer, 1993

Five users find most of the problems.

But only if there is something to use.

5
test users
84%
of problems found
Share of usability problems found as test users are added, rising steeply and flattening after about five users. 0%25%50%75%100%051015 test users
Found = 1 − (1 − 0.31)n, where 0.31 is the average share one user reveals | NN/g | CHI ’93

Dow et al., Stanford, 2010

Build several. Then choose.

Parallel prototyping led to better designs, more varied ideas, and more confident designers.

Sequential prototyping refines one design, v1 to v2 to v3. Parallel prototyping builds A, B and C at once, then picks the best. Sequential Parallel v1v2v3ShipABCPick the bestShip Dow et al., ACM TOCHI, 2010

What changed

AI cut the cost of a demo.

The gains land on new work.

A new build from scratch
55.8%
faster with an AI pair programmer
0%25%50%75%100%

Randomized trial: build an HTTP server in JavaScript. Shaded: 95% confidence interval, 21 to 89%.

Peng et al., 2023
A large codebase you already know
19%
slower, while it felt faster
Economists predicted
39%
ML experts predicted
38%
Developers predicted
24%
Developers estimated after
20%
Measured
19%
← fasterslower →

Randomized trial: experienced developers, their own mature repositories.

METR, 2025

New product development is new work.

Parkinson’s law of triviality, 1957

Memos collect opinions.
Demos collect bugs.

  • ACan we use the standard template?
  • BMove Background above Goals.
  • CTitle case for the headings?
  • DCould this section be a table?
Comments on a memo
  • 1Save took four seconds.
  • 2I couldn’t find undo.
  • 3Pasting an image crashed it.
  • 4This flow is great. Ship it.
Comments on a demo
Illustration | Effort flows to whatever is easiest to have an opinion about | Kamp’s bikeshed email, 1999

Demos in practice

Watch the room change.

iPhone, 2007. He didn’t describe scrolling. He scrolled.
IDEO, 1999. A new shopping cart in five days, argued with prototypes.
The marshmallow challenge. Tom Wujec on why prototyping early wins.
Macintosh, 1984. Out of the bag, and it speaks for itself.

Built into the culture

Where the demo is the deliverable.

“Demo or die.”
MIT Media Lab
Demo Day.
Y Combinator, since 2005
“Running code.”
David Clark, IETF, 1992
“Working software.”
Agile Manifesto, 2001

In fairness

Where the memo still wins.

Does the question have a runnable answer?
Yes
Build a demo.
  • Feasibility
  • Interaction
  • Performance
  • Will anyone use it?
No
Write a memo.
  • Strategy
  • Org design
  • Postmortems
  • Legal and financial

Write to decide what to build.
Build to find out whether it works.

Sources and further reading

Research

Risk, iteration and AI

Primary documents

Archives and reporting

“A memo is a map.
A demo is the territory.”

Will Murphy