Seatbelts
When people would ask me what I do for work, I would playfully reply: "I talk to computers."
It was true. Except I didn't literally use my speech. I needed to learn special syntax and control structures so that the computer could understand my intent. Those basic primitives alone took me far until I needed to co-author the intent with n humans and/or as the intent evolved. Version control, interfaces to hide complexity and holistic design allowed groups of people to continue balancing shipping software with ever growing complexity. In 2026, I still talk to computers; however, I've replaced writing syntax by hand with plain English language for a lot of my day to day tasks (full stack web software development).
There's a lot of discussion right now on the future of a software developer and zooming out further, the expansion of automation and AI across domains. I'll talk about the former and my experience as I can't speak to the latter and don't want to spread further alarmist or engagement farming takes. May the reader take this and other future postings as another data point.
Prior to large language models (LLMs), a developer didn't just write code. Some of the most time consuming tasks included research and building confident comprehension of a system. The surface area of a programming language is vast and holding all of its rules within working memory was not feasible. Sometimes I would find myself in so many Google searches to build understanding of a new language I was working with. Other times I would write meticulous notes and diagrams so I could batch questions to more senior engineers and verify my learnings of existing code and systems. After LLMs, the time for these tasks has been reduced as the answer to questions are more generally accessible along with an infinitely patient tool decomposing complexity of concepts. The latter is so valuable for me as someone that still deals with a good amount of impostor syndrome: a psychologically safe exploration sandbox to build confidence in a foreign piece of software.
The rate of change has moved incredibly fast. For example, at my last job working to support enterprise customers (2H 2024), I was still individually typing out each letter to apply the earliest efforts into what we now call AI Agents. At my job today, all of my engineering co-workers command an autonomous agent. In other words, machines are now writing the code for other machines to run.
Looking back, perhaps this will end up being my earliest sparklings of witnessed recursion in the singularity.
...
I do not know what the future holds but after my exposure so far, I'll say this for rest of the world watching this technological wave unfold:
The changes are very real and visceral.
The companies missed a key step: the public has little to no trust about intentions with such a powerful technology in a time where negative sentiment on technology and algorithms was already climbing high.
We are starting to feel the previously predicted geopolitical race: the general public's relationship with datacenters and mass surveillance cameras are among some of our first major decisions on how to proceed and apply the technology.
Should we assume these annecdotes of leverage increasing, feedback loops tightening and more shots on goal to deliver a desired outcome at a much cheaper rate are true, then our rate of progress has increased. Every day we are getting faster at getting closer to unlocking autonomous intelligence that matches or surpasses our species.
Organizational, policy and technological accessibility are some of the levers that control how we all end up feeling the rate of change above at some point in time.
It has never been easier to accumulate, synthesize and apply knowledge for your own goals.
Machines that can drive a growing number of outcomes have arrived. For now: how can I begin to have this conversation in my corner of the Earth?