Describing software and writing it are turning into the same job
Type "a button that says add $3 and a button that says withdraw $5, then show me my balance" and a tool built on GPT-3 returns a working React app. Nicole van der Hoeven, a k6.io developer advocate with ten years in performance engineering, argues that code and speech are converging from both directions. If a sentence can be a program, what should coders and writers each start doing now?
Get concrete habits, books and tools to stay relevant as code and writing converge
Nicole van der Hoeven, Developer Advocate at k6.io, argues that code is becoming more like human language while technical writing is becoming more like code, illustrated by a GPT-3 tool that builds a React app from a sentence. She gives coders and writers concrete habits to stay relevant: write expressive code, document your work, adopt developer tools for writing, learn in public and keep a personal knowledge management system.
“And the truth is that it's becoming way more important that your code be understandable by others than that you make it work in the first place.”
“Programmers rarely believe that they have to get it right the first time, but for some reason we writers do.”
What you will learn
Explain how programming languages climb layers of abstraction from binary toward human-readable language
Recognize how technical writing narrows from project plans to epics, user stories and Gherkin specifications that resemble code
Write more expressive code through clear naming, comments, atomic functions and peer review
Document your work with changelogs, release notes, readmes and contribution guidelines as part of the job
Apply developer tools such as Git, Markdown and static site generators to writing, and learn in public with a personal knowledge management system
Key takeaways from the speaker
Learn how to capitalize on the convergence of code and speech and make your career futureproof
Explore easy ways to branch out from speech into code, and from code into speech
Get up to speed on concrete tools that bridge the gap between speech and code
Details
Level
All levels
Format
Talk
Technical depth
Introductory technical
For
Early career, Mid-career, Career changer
Length
18:41
Published
June 2021
Questions this talk answers
In what sense is code moving toward speech?
Each layer of code, from binary and machine code to assembly, compiled and interpreted languages, is more abstract and closer to readable English. Tools based on GPT-3 now turn plain sentences into working JavaScript.
How is speech moving toward code?
Natural language already has strict hidden syntax, such as English adjective order, and technical writing becomes shorter, more structured and more specific as it moves from project plans to epics, user stories and Gherkin specifications.
What should developers do to stay relevant?
Write expressive code that others can follow, use peer reviews, and document their work with release notes, changelogs, readmes and contribution guidelines, as k6 developers do for every feature.
What should writers do to stay relevant?
Use developer tools such as Git, Markdown and static sites like Hugo, treat writing as a digital garden to refactor, and learn some code that they will actually use.
Why learn in public and keep a personal knowledge system?
Sharing what you learn brings help from others, helps people facing the same problems and leaves a record of career capital. Notes written for a future self who has forgotten everything help you remember and think deeply.
About this talk
Programming code and human speech are often seen as being positioned at different ends of the spectrum. At one end, code is terse, strictly logical, and limited. On the other end, speech is unbounded and often verbose. However, the gap between the two is narrowing. Programming languages are becoming more and more abstracted, and writing tools are becoming more and more action-oriented. How can we prepare for a future where coding and writing are the same thing?
Learn in public and keep your career capital (90 s)
Nicole van der Hoeven: Hi, everyone. I'm Nicole van der Hoeven. I'm a developer advocate at k6.io, and I've been a performance tester for several years. Being a developer advocate means being able to write code and also being able to talk about it. So this topic is of personal interest to me. Code and speech are often seen as being positioned at different ends of the spectrum. However, I think that they have more in common than is immediately obvious. We're going to go quickly through how code has been changing, then how speech has been changing, and then we'll talk about what we can do to adapt to these changes. And by the way, these slides are public and I'll share the URL at the end so you don't have to take notes.
Speaker 1: So on the one hand, there's code. Code comes in many forms, but this is the first program ever, an algorithm for an analytical engine that was created by the first ever computer programmer, a woman named Ada Lovelace. The interesting thing is that this is pure mathematics. Code is terse, strictly logical and limited. It has a vocabulary all of its own, and its syntax is pretty unforgivable.
Code has traditionally been in the domain of computer science and the stereotype is that it's incomprehensible to those outside of the know. And when we look at this precursor of code, we may be inclined to agree. Speaker 1: This is binary code. Binary code consists of zeros and ones and is used to encode data into a simple on/off or a yes or no pattern. By the way, it's not so important to understand all of this, but I just want to take us through a bit of the history of code.
Speaker 1: So each zero or one is a binary digit or bit. And then when we string together eight of them, we get unique representations of characters and we call that a byte. Speaker 1: Already we can see here a form of abstraction. Abstraction just means grouping things together to give them a collective name so that we can understand them better. And it's easier to refer to them. And it's something that humans are great at.
Speaker 1: This is machine code. Machine code interprets binary code in hexadecimal form. It's used to send instructions directly to a computer CPU. Speaker 1: And this is assembly code, specifically for a Motorola 68000 microprocessor. Assembly code makes machine code more functional, adding special keywords. So this is a huge jump in abstraction as Speaker 1: programmers at this level can already use words instead of just numbers or math.
Speaker 1: And the next layer is compiled languages. This is a snippet of a go code. At this point, the code is starting to become more intelligible, more human readable. Let me just scroll here. Speaker 1: There's still a structure there, but there's a distinct shift away from numbers and towards readable English words. Go is a compiled language, which means that all of this code is still translated into machine code before it's run, but it's a lot easier to understand at this level.
Speaker 1: And this is an interpreted language, specifically React JS JavaScript. We can see that this is much more abstracted again. Speaker 1: Now we're talking about account balances, states, buttons and clicking things that we already understand. Yet JavaScript is still translated to machine code at runtime. Speaker 1: So this is what we've got so far. You can call it a code abstraction cake of sorts. So we start with a layer of mathematics and physics, and we work our way up to interpreted languages.
Speaker 1: Code at each layer exists today, so it's less a chronological progression than just different ways to do the same thing. Speaker 1: And each new layer gets more and more abstract. That is closer and closer to human speech. So code is moving towards speech. But what about speech? Speaker 1: Speech is actually moving the opposite way towards coding. So take the phrase large snorting ancient red dragon.
Speaker 1: In this phase, the phrase, the dragon is a noun, the thing, and all the other words are adjectives. So they describe that noun that the dragon in some way. Speaker 1: But why don't we say, you know, red, large, ancient snorting dragon? Speaker 1: English speakers will probably say it sounds off, but what does that actually mean? And why is it that it sounds off?
Speaker 1: It turns out that there's this weird thing in English that when there are multiple adjectives in a row that refer to the same noun, they're always ordered according to the characteristic that they describe. Speaker 1: So large is a dragon size. The fact that it's snorting is a physical quality. Ancient refers to age and red is its color.
Speaker 1: In fact, adjectives in English are always written in this order. This is the full one from the Cambridge dictionary when there's more than one adjective for a noun. Speaker 1: This is an incredibly complex syntax. If we had a programming language that enforced this kind of complexity, it likely never again be used. Speaker 1: But native speakers of English just do this without even noticing, without having to memorize this list.
Speaker 1: So we often think of speech as unbounded and forgiving in syntax, but that's not always the case. Speaker 1: While literary speech often breaks grammatical rules and is celebrated for it, technical speech sticks to the rules. Speaker 1: To be able to compare speech with code, let's look at layers of written speech that we use to describe new product features in software.
Speaker 1: When we first start conceiving of a feature, we start with the most abstract level. Speaker 1: A feature usually starts with some sort of project plan, sometimes called a theme. Speaker 1: And a project plan is usually a pages long document describing in prose the reason for a feature or a group of features and giving a rough overview of how it might be implemented.
Speaker 1: And then the next level of abstraction down is epics. Epics are still not at the nitty gritty level, but at this level, we're already starting to see some structure. Speaker 1: In agile methodologies, epics answer a standard set of prompts. We're breaking up those pages into smaller sentences that are more actionable. Speaker 1: Next level down is user stories, and they're typically included in epics. They zoom in a little bit more on a feature.
Speaker 1: So here's a standard formula for a user story. As an account owner, I want to deposit and withdraw money from my account so that I can save and spend more. Speaker 1: The phrases as a I want to and so that are important here. Speaker 1: A likely offshoot of a user story is a behavioral specification.
Speaker 1: This is an example of one written in Gherkin, a business readable language. Speaker 1: The goal of Gherkin is to translate business requirements into a more logical structure that can serve as the basis for both tests and code. Speaker 1: Gherkin structure is less ambiguous, making it easier for developers to interpret what those requirements mean into programming languages.
Speaker 1: The words feature, scenario, when, and then are still prompts, but they're also keywords, kind of like the logical keywords, if, then, when, else that we see in code. Speaker 1: And as speech goes, it's way less abstract than project plans. Speaker 1: So this speech abstraction cake is upside down, and it starts with the most abstract layer, which is project plans.
Speaker 1: And as we go down through the layer, speech gets more structured, shorter in length, and more specific. Speaker 1: In short, it's getting more and more like code. Speaker 1: So here's one example. Speaker 1: Is this code or speech, a button that says add $3 and a button that says withdraw $5, then show me my balance.
Speaker 1: This is an example that exists in the intersection of code and speech. Speaker 1: It has the syntax of code, but it's human readable like speech. Speaker 1: So it's pretty easy to understand, even though we don't have any context. Speaker 1: So here's the context. Speaker 1: This is an app created by the company DeBuild.
Speaker 1: It's based on artificial intelligence called GPT-3 by the company OpenAI. Speaker 1: You type text that describes an app that you'd like to build, and GPT-3 builds a functioning React.js app. Speaker 1: So this is it being used and the buttons are being clicked. Speaker 1: And if you see down below, the code has been generated.
Speaker 1: That's actual JavaScript and pretty smart JavaScript at that all by an artificial intelligence. Speaker 1: So code or speech? Speaker 1: We've just gone from the description above to the functioning interface below. Speaker 1: Is that not what code does? Speaker 1: So code is moving towards speech. Speaker 1: I would argue that the sentences above are code, a new code, a kind of code that's written in natural language, but still code with a vocabulary and a syntax all its own.
Speaker 1: And at the same time, speech is moving towards code because those sentences are speech too. Speaker 1: They're immediately understandable, follow the rules of grammar. Speaker 1: They're even punctuated for like emphasis. Speaker 1: So on the right side, the code side, artificial intelligence is getting increasingly smarter and we're building types of code of higher and higher levels of abstraction.
Speaker 1: And because of this, we're able to give computers instructions that are closer to writing than the algorithm Ada Lovelace began with. Speaker 1: And then on the left side, the speech side, we're starting from a place of high abstraction already. Speaker 1: But working in tech and building software has caused us to be more and more precise with our language to the point where our descriptions of code may just become code themselves.
Speaker 1: So what can we do to remain competitive and relevant in the face of these changes? Speaker 1: Here are some concrete tools, techniques and areas of study that we can focus on to adapt to this new paradigm. Speaker 1: If you're a coder, write expressive code. Speaker 1: This meme pokes fun at the fact that actual programming takes up such a small slice of a coder's pie of work.
Speaker 1: It's an intentionally exaggerated example, but I think that there is a bit of truth in it. Speaker 1: And the truth is that it's becoming way more important that your code be understandable by others than that you make it work in the first place. Speaker 1: Because if you're the only one who understands how it works, then it's only a short term success.
Speaker 1: It only lasts as long as you stay on the project. Speaker 1: And after that, all bets are off. Speaker 1: As coders, we can pay more attention to the way the variables are named, yes, but also comments that we leave in our code, Speaker 1: not mixing levels of abstraction, keeping functions atomic. Speaker 1: This is why peer reviews are so important.
Speaker 1: If you give your code to someone else and they can't follow your train of thought, there's a problem. Speaker 1: It might not be a problem you have to solve now, but someone's going to have to do it at some point. Speaker 1: The second thing is document your work. Speaker 1: Many things can't be explained by just comments and code.
Speaker 1: And for those, you need to accompany your code with a more detailed explanation. Speaker 1: So the company that I work for maintains an open source load testing tool and our developers are great at taking responsibility for explaining their own work. Speaker 1: So at K6, they write documentation as part of a release. Speaker 1: Developers write documentation for every new feature that they build.
Speaker 1: They also answer questions in our public community forum. Speaker 1: They just treat it as part of their job because it is. Speaker 1: If you're a coder, make it part of your job to describe what you've done and why. Speaker 1: Write a change log and release notes, even if they're just bullet points. Speaker 1: Write readmes for repos, write contribution guidelines.
Speaker 1: It's your work after all. Speaker 1: You should be proud of it. Speaker 1: Learn how to write by reading what others have written and doing a lot of practice. Speaker 1: Here's a book that I found personally useful, and it's called Everybody Writes by Anne Hadley. Speaker 1: It's a good book to start with, and I also use tools like Grammarly, which check not just spelling or grammar, Speaker 1: but also tone and consistency and whether or not words I'm choosing are ambiguous.
Speaker 1: So if you're a writer, try using tools that developers use even for your writing. Speaker 1: Another book recommendation, Modern Technical Writing by Andrew Etter. Speaker 1: It talks about tools for writing that bring writing more in line with code with a goal of streamlining processes for both. Speaker 1: For example, we could store our writing on version control systems like Git, write in XML via lightweight markup language like Markdown, Speaker 1: and use static websites like Hugo, which is my personal pick.
Speaker 1: Using the tools that developers use makes the transition to code much easier, and you might actually start to see it creeping naturally into your writing. Speaker 1: Programmers rarely believe that they have to get it right the first time, but for some reason we writers do. Speaker 1: We should treat our work like a digital garden that's constantly growing and changing and refactor it like code.
Speaker 1: We should get into the habit of updating blog posts when we learn something new, Speaker 1: and think of writing as a perpetual work in progress rather than a once-off project with a clear end. Speaker 1: Learn to code, or if you already know how to code, get better at it. Speaker 1: Learning code doesn't mean you have to be a programmer.
Speaker 1: It just means knowing how to use it to make your work as a writer better. Speaker 1: I always recommend starting with something that you'll actually use, Speaker 1: because just like human languages, you'll forget any programming language you don't use regularly. Speaker 1: I'm a big fan of Coursera for learning a variety of subjects, and I highly recommend them.
Speaker 1: They're free. Speaker 1: You can also pay to get certified there. Speaker 1: And here are some courses I've found useful on other sites, and all of them are free. Speaker 1: For both of them, for both writers and coders, here are two takeaways. Speaker 1: The first is learn in public. Speaker 1: Talk about what you're learning, and especially if you're not an expert yet.
Speaker 1: Share it, whether it's on social media, or a blog post, or a video, or a conference, or on your own website. Speaker 1: Not because we're self-centered and we want to show off, but because we just want to have a record of everything that we've done. Speaker 1: Why should we throw away all that career capital?
Speaker 1: Here are three of my favorite public learners. Speaker 1: Chris Sean has hundreds of videos on YouTube documenting his journey into software development as a complete beginner. Speaker 1: Josh Branshaw has a GitHub repository called TIL, that stands for Today I Learned, where he just posts one thing a day that he's learned every day. Speaker 1: And years later, he has a repository of over a thousand little snippets.
Speaker 1: Sean Wang blogged his blog. Speaker 1: Sean Wang blogged his blog for a developer enough that he found he had enough to write a book called The Coding Career Handbook. Speaker 1: In my personal experience, learning in public helps you learn faster because people chime in and help you along while helping others who may be struggling with the same thing and leaving a record of where you've been for future employers or for yourself.
Speaker 1: Secondly, build and maintain a personal knowledge management system. Speaker 1: I think that's a great idea. Speaker 1: Take notes on your daily activities, on things you're learning, as if you're talking to a version of yourself that has forgotten everything, because I guarantee that you will forget things years down the track. Speaker 1: I've personally used tools like Obsidian, Roam Research, and Notion to remember and think deeply about topics that interest me or ones I'm learning.
This on the left is a graph view of my notes on Obsidian. You can check it out. I've published it at notes.nicolevdh.com. It's messy, so don't expect something polished, but it just helps me make sense of this crazy world of tech. Speaker 1: The thing with change is that everyone wants it, but nobody wants to change. The convergence of code and speech can be daunting and is causing many of us to rethink the way that we work.
I hope that the tools and concepts I've mentioned today will help some of you prepare for this convergence and look at it with excitement rather than fear. Speaker 1: Thanks for hearing me out, and if you want to check out any of the resources that I mentioned, as promised, here are my slides. You can find a copy of them on slides.nicolevdh.com.
And if you go there, you can also, here's a tip, you can click S on your keyboard and you'll see the speaker notes just so you can hear the little Nicole in your head going through these slides. Feel free to reach out to me on all of these social channels as well. And thanks again for listening.
Transcribed automatically; names checked against the speaker record. Select a time to jump there.
Article based on this talk
Why coders must learn to write and writers must learn to code
Nicole van der Hoeven traces how code and human language are meeting in the middle, and what to do about it.
In short
Code and speech are converging. Nicole van der Hoeven, Developer Advocate at k6.io, shows that programming languages keep climbing toward readable English while technical writing keeps tightening toward the structure of code, to the point where a GPT-3 tool can build a working app from a sentence. Her answer is practical: coders should write expressive code and document it, writers should adopt developer tools and learn some code, and everyone should learn in public and keep a personal knowledge system.
Key takeaways
Learn how to capitalize on the convergence of code and speech and make your career futureproof.
Explore easy ways to branch out from speech into code, and from code into speech.
Get up to speed on concrete tools that bridge the gap between speech and code.
Code is climbing toward human language
Van der Hoeven opens with the first program ever written, Ada Lovelace's algorithm for the analytical engine, which she describes as pure mathematics: terse, strictly logical and unforgiving. From there she walks up what she calls a "code abstraction cake": binary digits grouped into bytes, machine code in hexadecimal, assembly code for a Motorola 68000 that adds keywords, compiled languages such as Go, and finally interpreted languages such as React JavaScript, where the code talks about account balances, buttons and clicks.
Her central idea is abstraction, which she defines simply as grouping things and giving them a collective name so they are easier to understand and refer to. Each layer is more abstract than the one below it, and therefore closer to the way people talk. She is careful to note that this is not a timeline: "Code at each layer exists today, so it's less a chronological progression than just different ways to do the same thing." (4:07)
Speech is stricter than it looks
The surprise of the talk is that speech is moving the other way. Van der Hoeven uses the phrase "large snorting ancient red dragon" to show that English adjectives follow a fixed order by size, physical quality, age and colour, a rule she shows in full from the Cambridge Dictionary. It is a syntax so complex that no programming language could get away with it, and yet "But native speakers of English just do this without even noticing, without having to memorize this list." (5:20)
She then builds a second, upside-down cake from the way software teams describe features. A project plan is pages of prose. Epics break it into structured prompts. User stories follow a formula ("As an account owner, I want to deposit and withdraw money from my account so that I can save and spend more"). Behavioural specifications in Gherkin use words such as feature, scenario, when and then that act as prompts and keywords at once, much like if, then and else in code. As she puts it, "But working in tech and building software has caused us to be more and more precise with our language to the point where our descriptions of code may just become code themselves." (10:22)
Where the two meet: an app from a sentence
The meeting point is a demo from Debuild, built on OpenAI's GPT-3. A user types a description, "a button that says add $3 and a button that says withdraw $5, then show me my balance", and the tool produces a functioning React app with the JavaScript visible beneath it. Van der Hoeven asks whether that is not exactly what code does, and answers: "I would argue that the sentences above are code, a new code, a kind of code that's written in natural language, but still code with a vocabulary and a syntax all its own." (9:24) The same sentences are speech too: grammatical, punctuated and understandable without context.
What coders should do: be understood
For developers, her first piece of advice is to write expressive code. Naming variables well, leaving useful comments, not mixing levels of abstraction and keeping functions atomic all matter, which is why peer reviews are so important. "And the truth is that it's becoming way more important that your code be understandable by others than that you make it work in the first place." (11:12) Code only one person understands is a short-term success that lasts as long as that person stays on the project.
Her second piece is to document your work. At k6, which maintains an open source load testing tool, developers write documentation for every new feature as part of a release and answer questions in the public community forum. "They just treat it as part of their job because it is." (12:36) She recommends changelogs, release notes (even as bullet points), readmes and contribution guidelines, along with practice, the book Everybody Writes by Ann Handley and tools such as Grammarly that check tone, consistency and ambiguity.
What writers and everyone else should do
Writers, she suggests, should borrow developers' tools. She recommends Modern Technical Writing by Andrew Etter and describes storing writing in Git, writing in Markdown and publishing with static site generators such as Hugo, her personal pick. She also urges writers to borrow developers' attitude to iteration: "Programmers rarely believe that they have to get it right the first time, but for some reason we writers do." (13:59) Writing becomes a digital garden, refactored like code. And writers should learn to code, starting with something they will actually use, through resources such as Coursera.
For both groups she offers two habits. The first is to learn in public on social media, blogs, videos, conferences or a personal site, citing Chris Sean's YouTube journey, Josh Branchaud's Today I Learned repository of over a thousand snippets and Shawn Wang, who blogged enough to write The Coding Career Handbook. "Why should we throw away all that career capital?" (15:27) The second is a personal knowledge management system, with notes written for a future self who has forgotten everything, using tools such as Obsidian, Roam Research and Notion. She closes with a reminder that "The thing with change is that everyone wants it, but nobody wants to change." (17:20)
Watch the talk
The Convergence of Code and Speech, Nicole van der Hoeven, about 19 minutes: https://www.womentech.net/video/convergence-code-and-speech
About the speaker
Nicole van der Hoeven is a Developer Advocate (Performance Testing) at k6.io and a performance engineer with ten years' experience in breaking software and learning to build it back up again. She has worked as a performance testing consultant in Australia and Europe and enjoys drawing on fields such as economics, video production, linguistics and productivity to reframe load testing concepts.
Questions this talk answers
In what sense is code moving toward speech?
How is speech moving toward code?
What should developers do to stay relevant?
What should writers do to stay relevant?
Why learn in public and keep a personal knowledge system?
Drafted with AI from the talk transcript. Editor review pending.
Frameworks and ideas mentioned
Agile epics and user stories
User story formula (As a, I want to, so that)
Gherkin behavioural specifications
Learning in public
Personal knowledge management
Digital garden
Resources mentioned
Everybody Writes by Ann Handley
Modern Technical Writing by Andrew Etter
The Coding Career Handbook by Shawn Wang
Grammarly
Coursera
Git
Markdown
Hugo
Obsidian
Roam Research
Notion
Josh Branchaud's TIL (Today I Learned) GitHub repository
Chris Sean's YouTube videos
notes.nicolevdh.com
slides.nicolevdh.com
Cite this talk
Nicole van der Hoeven. "The Convergence of Code and Speech." WomenTech Network, 2021. https://www.womentech.net/video/convergence-code-and-speech