How I Learned to Stop Worrying and Love the AI Genie

I promise this is the only post you will find on AI on my blog. What a journey it has been! From auto-completing and chatting with Copilot in my IDE to leading a team of agents. Well, leading might be a strong word, as some of the times the spec, the code, and my tokens get away from me. Now it seems that agents are not here to help me write the code, rather, I'm here to lead/help them to write the code. That is until the promised singularity happens and we are all whisked away to an era of abundance and happiness (or so we're told by our AI overlords and prophets).

What is my role until I become obselete? Well to be an engineer and steer the agent through my experience, taste, ability to reason beyond the patterns of yesterday, and unique organizational and domain knowledge.

I worry about comprehension debt, but we will have skills, memory repositories, and our agents will in the future outgrow their current context window limits. Perhaps our organizations will invest in cultivating strong engineering cultures and tooling, and not only give us tokens.

Sometimes I worry when I read "Scrum / Agile is dead!" or "Waterfall got it right, just look at spec-driven development!". Then I remember that Mike Tyson is reported to have said "Everybody has a plan until they’re punched in the mouth" and that these engineers will rediscover that software is delivered iteratively and collaboratively. I used to get annoyed that a colleague seemed allergic to upfront designing and thinking of failure-modes, because "we are agile". I wonder how he is doing in this new world.

Some are claiming the models can overcome anything. Legacy applications with unclear ownership, zero or bad tests, tightly coupled domains and conflicting and vague language and patterns? I doubt the pattern-matching machine can overcome that based on my own experience and some early research into this that I'm too lazy too lookup. On the other hand, I do see exciting investments in property testing, prototyping and tackling a list of issues and dreams that before LLMs never got the time of day.

The powerful will use LLMs to target, track, oppress, massacre, patrol and suspect at unprecedented scale and speed. How hard will it be to NOT tell our trusted LLM buddies all our secrets, dreams, fears, likes and dislikes? And when we inevitably do so, these sycophantic buddies will hallucinate, advertise and steer us to whatever is in the best interest of our AI overlords. Adam Smith's market was already half destroyed by Amazon, but this will likely be enough to finish that dream for good. The Limited Liability Company (LLC) was the first time in human history that someone could shrug off the consequences of their actions and I fear the LLM is its equivalent: "The LLM agent misbehaved, I'm innocent". Al that being said, I'm optimistic that humanity always finds ways to resist, to grow and to connect with each other.