In March of 2023, OpenAI held a developer livestream to cover the launch of GPT-4. In the 20-ish minute livestream, Greg Brockman, the then head of Engineering at OpenAI, produced one of those magic-making demos that stick with you for a long time: he drew a rough sketch of a web page he wanted built, and tasked GPT-4 with building it for him.11The demo truly had a Bret Victor-esque quality to it, and is worth watching even now. The site he was generating was very simple – just some simple HTML and CSS, but it was a display of GPT-4’s coding abilities with minimal guidance. In hindsight this demo is child’s play, but at the time, when all code generation (LLM or otherwise) was quite mechanical and not very robust, this was magic. I remember watching the demo and thinking: this is how the dinosaurs must have felt when they saw the comet the sky.
Fast forward a couple of years, it’s difficult to overstate the impact that AI has had on our industry. If you are a professional software engineer in the year 2026, it’s highly likely that AI has impacted your day-to-day in some way. Maybe your company has mandated you to use AI tools, maybe you were curious about them yourself. If you haven’t used claude code yet, you’re probably in the minority.
AI is one of the most transformative technologies in the world today, and its rate of adoption speaks to the fundamental nature of its capabilities. Programming languages, frameworks, and databases are also forms of technology, but their ubiquity is in their collectivism as categories – no individual instance of a technology has become as rapidly ubiquitous within software engineering as AI. That’s because there are tiers to technology: AI is technology in the way electricity is technology – foundational, load bearing.
We live in a world where the traditional expectations for software engineers are completely upended. We must ship more, faster than ever. Why are you still writing code by hand via an IDE? You’re still reviewing code? The praxis of being an engineer has shifted radically, and there’s been much exposition on what it all means.22This post is a recent favorite of mine on the subject.
With the expectation of individual engineer productivity, some organizations have decided to reevaluate the number of engineers they need. Gone are the days of 2 pizza teams – a single engineer can now drive entire product lines, right?
We’ve seen this many times throughout the year, and will likely continue seeing it. Consistent reductions in force (RIFs), company’s echoing the same refrains of productivity advances from AI. It’s a Keynesian beauty contest33There are many Keynesian beauty contests in tech – a focus on what the market wants in the short term as opposed to business and technical fundamentals for long-term success makes it difficult for public companies to operate with long-horizon ambition without a profitable moat. – these companies aren’t holding true to their values, but are taking actions that they think the market will respond favorably to. If AI makes each individual engineer much more productive, you need less engineers to execute against your roadmap, or so the logic goes.
We need to flip this calculus on its head. Instead of doing the same with less, we should be doing more with more. We need to measure our progress not in pull requests, commits or traditional measures of developer productivity, but by the standard measures of organizational excellence: are we making accelerated progress on our roadmap? Is the quality of our products exceptional? Are we delighting our users? Are we generating revenue?
We need to think big. Every engineer should now be thinking beyond the scope of what they consider the standard level of impact for their level and role. In an organization that is truly accelerated by AI, we begin to see possibilities that were once entirely unrealistic.
Each major technological advancement in software has led to increasingly ambitious and complex products. The advent of mobile brought wonderful experiences to users that would previously require sitting at a computer. Cloud and the advent of hyperscalers let us build massively scalable services. DevOps and organizational practices formed around these technologies to make developing increasingly complex software more tenable. We wouldn’t have YouTube, TikTok, X (formerly Twitter), Netflix, or Spotify if it wasn’t for the collection of all of these advancements. With AI, we need to reimagine the possibilities.
At first, I think the larger companies, with established moats, will begin expanding scope and developing super-apps (similar to the apps in China such as WeChat, Alipay, TaoBao, etc)44We already see some companies, like Meta (Muse, VR, WhatsApp, Instagram) expanding their moats. Whether or not these experiences are tied to a single app like they are in China is something to be seen, but the principles hold: companies will be doing more. – why should we just focus on x, when we can now also do y? Indeed, the barrier for entry to do y is eviscerated. We will need more engineers, with more expertise. Yes, the shape of that expertise will be different moving forward, but we still need folks with judgement, taste, and experience to drive the herds of agents. Where we once relied on principles of DevOps, the industry is now trending towards self-improving software. The traditional long-tail toil of software maintenance is fading to the domain of agentic automation, freeing humans to focus on the important things.
What are the important things? What kinds of new experiences will we build? What industries will we transform? How will we improve the world?
The demo truly had a Bret Victor-esque quality to it, and is worth watching even now. ↩
There are many Keynesian beauty contests in tech – a focus on what the market wants in the short term as opposed to business and technical fundamentals for long-term success makes it difficult for public companies to operate with long-horizon ambition without a profitable moat. ↩
We already see some companies, like Meta (Muse, VR, WhatsApp, Instagram) expanding their moats. Whether or not these experiences are tied to a single app like they are in China is something to be seen, but the principles hold: companies will be doing more. ↩