Feeling down — anxious about this future? That's ok! You're not alone and it's not all doom and gloom. Check out my follow-up article, "Nervous ChatGPT Will Take Your Programming Job?".

Update (4/26/23): This blog post was featured in Business Insider "The end of coding as we know it"

Update (4/21/23): A video of a fully autonomous AI agents created python code. Literally replacing programmers. Not that smart, but it shows the how possible this is TODAY, well ahead of schedule.

Update (4/17/23): GPT-4's abilities are mind-blowing and only strenghten what I've said here. https://www.youtube.com/watch?v=qbIk7-JPB2c (29:30 mark). Anyone who thinks GPT is just absent-minded text completion needs to take their heads out of the sand.

Update (4/3/23): for a counter response to this article, please see Don't worry chatgpt will not replace all programmers on AI Blog.

Update (3/5/23): if I can't convince you, maybe this article from 2017 where Andrej Karpathy (director of AI at Tesla/OpenAI) will.

In this article, I chart a course on how generative AI (GAI) like ChatGPT will supplant software engineers within a decade. The predictions are discretized into 5 phases with the overall trajectory trending toward total takeover. But first, a brief foreword.

Foreward — Addressing AI Fallacies

Human predictions are notoriously nearsighted, especially in regard to technology:

• In 1903, the New York Times predicted it would take humans over a million years to fly. Nine weeks later, the Wright Brothers did just that.
• That same year, a prestigious bank president said of investing in Henry Ford "the horse is here to stay but the automobile is only a novelty — a fad.
• Regarding the radio, "the wireless music box has no imaginable commercial value. Who would pay for a message sent to no one in particular?"

Yet here we are a century later, beholding the AI equivalent of the automobile and deluding ourselves into thinking it's somehow different this time. The car's just been invented, and we're saying it'll augment our horses with wheels.

Fallacy 1: AI will only replace script kiddies

Overlooking the obvious point that an industry without newbies will die in a generation, no earthly developer, junior, senior, or L9 mega architect is as proficient and broadly qualified to write software as AI. Can you claim to do all or any of these Day 1 achievements of ChatGPT?

1. Pass coding interviews at several top companies
2. Architect systems and database tables from user stories in seconds
3. Understands and translate between most languages (spoken and coding)

On top of this, ChatGPT4 is only weeks away and is now at Microsoft scale.

But look — its code has bugs!

News flash — world-class developers also generate code with bugs. The difference is that our cognitive abilities aren't growing at a geometric rate. When I first saw generative AI (GAI) create a website in Python, then translate it to Rust, my obsolescence was patent. Yet, the majority's mass reaction to some minor bug or out-of-place comment is "see look — it can't do what I can". How delusional do we sound? This is confirmation bias, plain and simple — akin to concluding that the first automobile would never replace carriages because the roads aren't wide enough.

Fallacy 2: AI is here to augment, not replace

Engineers are myopic and don't realize they constitute a massive cost center that management is salivating to slash. Putting aside that OpenAI has openly stated its intent to replace developers, consider how expensive we are. In 2021, there were 4.3 million developers in North America with an average salary of over $100k. Throw in benefits, taxes, and other employee expenses, and that conservatively comes to six-hundred billion dollars. You're telling me a guy like Bezos, who thinks bathroom breaks are too expensive, wants to keep us around? And once a few companies go this route, the rest will fall like dominoes to avoid being priced out.

Beyond just cost savings, there's a reason OpenAI has programming jobs in its crosshairs — it's the easiest industry to replace. Even though ChatGPT passed law and business school exams, those areas are less accessible to AI. For example, the law involves physical evidence, courtrooms, and paperwork, which would require robotics, a field wherein making a pizza is a breakthrough. And due process involves red tape, jury selection, appeals… things rate-limited by human time scales. By comparison, software engineering is entirely digital, exponentially scalable, and always behind schedule. It's the perfect target for a hostile takeover.

Predictions

Phase 0: The Prototypes (Q1 2023)

At the time of writing, we are in the prototype stage. This stage is characterized by nascent, browser-based AI tooling. OpenAI has captivated the world with DALL-E and ChatGPT and Google is responding with Bard. The majority of interactions happen through the web browser, although ChatGPT extensions are blossoming. Tools like Codex and CoPilot are already staking a foothold in the IDE, but major adoption is not worldwide. ChatGPT is still unstable and these platforms need productionized. However, this will soon be rectified with Microsoft and Google entering the fray (note — this happened in the few weeks since I began composing this article).

Job Loss Prediction: 2%

The industry in early 2023 has already endured massive tech layoffs, and GAI tools are quite immature. Management is still getting a grasp on them, and we're in a collective honeymoon phase. Although it is too early to replace most engineers, GAI is already disrupting the ad business. Yahoo is laying off 50% of its ad tech, and Google's search business is majorly threatened. Thus, I predict layoffs of engineers will continue through 2023, but they are incidental — a reaction to, rather than a consequence of.

Phase 1: Scale and IDE Infiltration (Q2 — Q4 2023)

CoPilot, Codex, and other IDE GAI tools will gain widespread adoption. For example, in a Java/Spring project, swaths of boilerplate code would be easy to generate:

Add a CRUD Rest API for managing user objects

This would result in the autogeneration of controller, service, and Hibernate modules. The main difference between Phases 1 and 0 is the amount of technical context maintained by the IDE. In phase 0, we'd have to instead prompt with something like

Show me an example of a REST API in Spring framework for CRUD on a user object. Include examples of Hibernate objects.

And then copy/paste the results ourselves. The productivity boost of having project context within the IDE will be immense. New project creation will also occur in the IDE and replace scaffolding tools like Spring Initializer.

Job Loss Prediction: 5%

The damage would be worse if 2023 wasn't already affected by massive tech layoffs and slow-moving industries (eg. banking) will be slower to adopt. GAI will first invade the initiative through its integration with the Microsoft tool suite. Its adoption otherwise will be marred by regulations and risk aversion of an industry stuck in the past, where on-premise solutions are the norm and "microservices" is still a buzzword.

Early adopters will otherwise see massive gains in productivity and accelerating engineering velocity. This might be the tailwind to eventually push us out of the recession, but in the short-term job growth is unlikely. Especially as the GAI-augmented engineering staff hit targets and deadlines like never before. Apart from companies like Google whose ad-revenue business will be rendered by GAI, most of us are safe. But the honeymoon of augmentation will be short-lived

Phase 2: Advanced IDE Tooling and Consolidation (1 — 2 yr)

Phase 2 will organically emerge from Phase 1, as IDE tooling gets more sophisticated. The entire codebase will provide context to the AI, which will then be able to make project-wide suggestions like:

Rebuild this Angular 6 project in the latest version of React

Add 100% unit-test coverage to the project

Refactor all model classes into a separate library using Gradle and Java 17

Change the security model from OAuth to SAML

This will drastically improve legacy codebase maintenance and migration. As millions of engineers apply such changes, usage patterns will emerge and be used to train the next generation of tools. As the traditional IDE becomes a mere vessel for AI, such suggestions will be auto-applied. For example, why wouldn't every codebase get 100% unit-test coverage? In fact, it should happen automatically in the background as we code.

This period will be known as the great consolidation. Long-held competitions between frameworks will be decided by which is the most AI-friendly. For example, imagine one could safely upgrade a legacy JS codebase to React with a single button click, but doing the same migration to Vue would take a week. Even if there are viable reasons to choose Vue, ultimately frameworks that adapt quickest to AI integration will prevail.

Such an arms race will occur in programming languages as well. For example, imagine you inherited a slow-performing code base written in Python. Your IDE suggests it be translated to Rust. You click "yes" and redeploy, and the app is suddenly 10X faster. The same goes for protocols — why bother with HTTP when everything can easily be gRPC? Let's get off of FTP and .md5 while we're at it.

Finally, Phase 2 will also bring the advent of the AI-CD pipeline. It's not difficult to imagine a CD pipeline that can learn deployment patterns. For example, an AI that tweaks Kubernetes configuration based on traffic daily. Or even goes as far as to rebuild the application in different tech stacks to optimize performance and server cost.

Job Loss Prediction: 25%

In Phase 2, the engineering purge truly begins. Whether that is due to augmented engineers being overly productive, or that AI has replicated usage patterns and manages code without them, the consequences are the same. The job loss hits hardest for:

• Low performers or AI-holdouts
• AI replaces generalists firsts. Web is the first domino, followed by mobile. Specialists like virtual reality or game developers are still safe.
• Tech companies whose main cost center is engineers. Elon's rending of Twitter portends the future of the tech industry.

Phase 3: SaaS and No-Code (2–5 yr)

By 2025, the digital glacier of Phase 2 has left in its wake a homogeneous software landscape. Legacy codebase bases have been overhauled or replaced. Certain frameworks and programming languages have emerged as the clear winners of the great consolidation. Active codebases meet unprecedented standards of test coverage, security, and standardization. Manual documentation is an anachronism. And all the while, AI has been learning the usage patterns of its IDE-augmented users. All this has set the stage for the most important inflection point in software history — AI now fully comprehends and operates at the level of business rules.

Phase 3 will be characterized by a Cambrian explosion of no-code AI tools, which enable non-technical users to build applications. Chat prompts now focus on the business logic in a Q&A fashion. For example, imagine a hypothetical food delivery service.

Q: What is the criteria for a customer to place a $200 food order? A: Customers purchasing a $200 order must have at least $75 of active credit in our system and have placed 3 or more orders in the last year.

Q: Please add a rule that customers cannot place more than a $200 order with 50 radial miles from their home address. A: That rule has been added — here is documentation of other rules you may be interested in.

Coding in an IDE will die off. In fact, the "code" is nothing more than an implementation detail, forgotten more with each passing day. The sole focus of these products becomes the human-AI interaction. Special-purpose wares for every facet of the stack emerge, for example, AI in the DevOps pipeline that specializes in blue-green deploys.

Q: Deploy the $200 rule change to the Canadian market first — if successful deploy in US over the next 3 months. A: I've configured a deployment model for this feature to rollout in south western territories first. If the new rule results in greater than 5% user interaction in February, I will apply a phased rollout in the US market beginning on March 22. Here is a tracking report. Would you like to be alerted at milestones?

Sites like GitHub will look start to resemble Wix, as they compete to court the new generation of non-technical users.

Job Loss Prediction: 75%

At this stage, the full-stack generalist is gone, as well as the product team and mobile developer. Niche fields like video gaming and medical software are now under threat. Developers transition to other engineering fields like robotics and biotech. All of this commences the age of the super SME (subject matter expert) — non-technical users with deep business knowledge who build businesses directly through AI.

Phase 4: AI Native and Domain Dominance (5–10 yr)

Phase 4 is the most interesting so far. AI-driven project management has led to a reduced and homogenized code ecosystem. As code is no longer maintained by humans, there's no need to optimize it for cognitive complexity. Unit tests, documentation, and design patterns die off — relics of a bygone era. The underlying code is a hidden detail, as inconsequential to its users as the firmware of a handheld calculator. Sure, the code could be rendered in Python, Java, or any other language, but to what end?

"The most amazing achievement of the computer software industry is its continuing cancellation of the steady and staggering gains made by the computer hardware industry." ~ Henry Peteroski

As AI-managed code converges on only the most performant tech stacks, the AI-Native era begins. New hardware and compilers are designed for AI operators. Runtime code is executed in novel assembler languages. Unshackled by paltry human cognition, design patterns are abandoned in pursuit of mechanical sympathy. Nvidia, Intel, and other manufacturers take full advantage of this. Any remaining human developers are pushed out, as the abstruse instruction sets of AI-native code may as well be written in ancient cuneiform.

Job Loss Prediction: 95%

While lumbering late-adoption industries like banking may have held out through phase 3, they can no longer afford the cost center of human beings; it's now shed or dead. The last vestiges of human developers are purged from the industry. Apart from AI research and the archaic COBOL programs running nuclear reactors, human coders are now an outmoded novelty, like the horse-drawn carriage rides in Central Park.

Phase 5: Heat Death (10+ yr)

Software is now unrecognizable. Business rules are maintained by AI-powered SMEs. AI-native standards are globalized. Code no longer resides in static repositories — it is ephemeral and dynamic. Every day, new AI tools and products appear. They are immediately plugged into a distributed network of programs and digital business is transformed.

Imagine your future self owning a digital business. You create and maintain it through an AI-powered design platform. A virtual fleet of AI makes daily improvements and suggestions. Performance and security enhancements happen automatically and unseen. Each time you log in, you're prompted about new features. Since AI completely understands the business domain, it can make domain-level suggestions.

AI: Your competitors are adding a buy now pay later option, would you like to incorporate this?

You: Yes — please deploy it to half of my US users.

This has all the makings for an arms race, as AI products compete to manage your business. Nefarious actors will also utilize and leverage AI to their own ends. The world jumps to hyper-speed — digital transformations arrive years ahead of schedule on the tailwind of AI.

Job Loss Prediction: 99%

By now humans have settled into their supplementary role as AI shepherds, or sheep if things don't go well. They've borne witness to a total transformation of the digital substrate. There will still be hobbyist coders of course, but their work will be of little consequence… something akin to oscilloscope art.

But all is not lost. While these predictions are grim, as with any industrial revolution, people tend to bounce back and thrive. After some short-term pain, and putting aside the possibility of total annihilation, what awaits humankind? A world where one machine can do the intellectual grunt work that used to preoccupy ten thousand. What will we do with it?

Level Up Coding

Thanks for being a part of our community! Before you go:

• 👏 Clap for the story and follow the author 👉
• 📰 View more content in the Level Up Coding publication
• 💰 Free coding interview course ⇒ View Course
• 🔔 Follow us: Twitter | LinkedIn | Newsletter

🚀👉 Join the Level Up talent collective and find an amazing job