Principles of Developer Experience Ray Tsang @saturnism James Ward @_JamesWard

Proprietary + Confidential ex·pe·ri·ence /ˌikˈspirēəns/ noun practical contact with and observation of facts or events

Setup / configure a service via command line or UI Use a library or API Read documentation / learn Troubleshoot problems Make something work. Anything! Developer Experiences

What is Good Developer Experience?

Bad

Good

node-v12.14.1-linux-x64.tar.xz How do I extract this?? Frustrating

Easy

Proprietary + Confidential Developer Experience is...

It feels good It feels nice It feels horrible It feels difficult It feels ...

No content

"If your application starts up in 2 seconds locally, but it takes 21 seconds to start in the Cloud…"

"It depends"

"It depends on what is more productive"

Value I can reach my goals, e.g. finishing the project I can get more than I expected DX = Value / Time Time I can reach my goals in shorter time I can finish a tutorial with fewer steps

Low Fewer steps in tutorial, but didn't learn anything useful High More steps, but accomplished more goals than otherwise possible High Spend little time, and everything worked! ZERO Spend a lot of time, and nothing worked It's a Ratio!

Productivity is measurable (But not in loc)

Brainstormed a lot of things that we feel is good Developer Experience

- Accomplishing something in a reasonable amount of time. - Try your experience from scratch, just like a normal user would - Initial success in 5 minutes - Fail fast with meaningful directional error messages (i.e. how do I resolve this error) - Getting started docs paired with very detailed reference docs - Layered approach - If a how-to references a feature / function then it should link to the detailed reference doc for that - A workaround is a bug - Not personal taste, but measurable productivity - Implement the right pattern the first time, and peel back abstractions - Highest abstractions first, - Local dev first & local testing - Incremental learning / No forward references - Guarantee success - Users don't remember details, they remember how they felt - Aim for a positive emotional reaction - Abstractions should not prevent learning - Users goal is to learn - Magic is frustrating - "Do this because of this" - Each step should be meaningful - I can take this & apply it somewhere else - Am I doing this to learn something or to satisfy the system - The DX needs to be consistent with the industry best practices - For example: semver, repeatable build, dependency - Docs must be versioned - Bring the beginner's mind - Assume the mind of your target audience - Minimality Rule - Don't make things hard for no reason - Do the simplest thing possible to achieve a goal/purpose - Make every step is minimal of a change as possible - Every part of a step should be meaningful - DRY - Avoid redundant - Information architecture should be feature oriented - Content types are: Quickstart, Feature Docs, API Docs - Quickstart - Feature 1 - Sample, tutorial, link to API Docs - Feature 2 - Sample, tutorial, link to API Docs - API Docs - Don't label with "How-to Guides" / "Tutorials" / "Samples" instead just list the feature names - Product overview should list the headline features - Typography matters - Documentation should build trust - Reflect reality - Don't say do "x" to achieve "y" if it's not true - Docs should not be used to smooth over product gaps & oddities - You can't fix the confusing overlap between Cloud Run and GAE by saying one is for "services" and one is for "apps"

Proprietary + Confidential dx.reduce(toPrinciples).size == 4

Before Principle 1 - Respect developer knowledge and goals After

Principle 1 - Respect developer knowledge and goals

"Was this experience consistent with the way I do things? yes or no" "Did you learn something useful? yes or no?" "Did you solve your problem, accomplish your goal? yes or no" Respect developer knowledge and goals Measurements - Principle 1

Before Principle 2 - Do the simplest thing that could possibly work After

Principle 2 - Do the simplest thing that could possibly work

How many of these to accomplish a goal: ● unrelated tasks ● required prerequisites ● steps ● snippets copied ● Google searches ● docs Do the simplest thing that could possibly work Measurements - Principle 2 ● CLI commands ● clicks ● context switches ● mistakes ● wasted effort ● decisions

Before Principle 3 - Learning should be incremental . . . After

Principle 3 - Learning should be incremental docs.docker.com/get-started

"Did you find the relevant information easily and quickly?" "Did you find what you were searching for in the first search result?" "Do you know where to go next to continue learning?" Learning should be incremental Measurements - Principle 3

Before Principle 4 - Wasted time is a waste After

"Did this take longer than it should have?" "How much time did you feel you needed vs how much time did you actually spend?" "How much time do you feel you've wasted?" "Which step prevented you from accomplishing the task?" Wasted time is a waste Measurements - Principle 4

Principle 4 - Wasted time is a waste Principle 3 - Learning should be incremental Principle 2 - Do the simplest thing that could possibly work Principle 1 - Respect developer knowledge and goals

But How?

Friction Logs

Subjective

Respect developer knowledge and goals The use/support of industry standard tools (Gradle, IntelliJ) is nice. There was some bespoke setup. Do the simplest thing that could possibly work There wasn't a good starter project that I could find / reference. Learning should be incremental Resolving my issues took significant time on StackOverflow and tinkering with the build. Wasted time is a waste It took much more time than it should have to get a simple build working.

Collecting Metrics

No content

What does it mean to be OK? Why isn't it excellent?

In addition to open feedback, ask the questions you want answers to. Make it easy for users to provide the feedback.

Other Ideas

Apply the principles from start Work with UX, PM, Docs, Libraries...

Proprietary + Confidential Questions?