You and I have an agreement.
If you send me slop, I will just start prompting you.
If you don’t want me to prompt you, as is probably the case, then your communication with me should be two-sided: On the one hand, I give you my attention, and on the other, you give me something of value in return.
This has been the defining human tradition for all of recorded history, and many millennia worth of prehistory besides. For example:
You yell “tiger!” in my direction.
I listen.
In return, I learn of a nearby tiger.
In the future, if I ever spot a tiger behind you, I will yell “tiger!” in your direction, because you have become my friend. Or at least, a useful tiger-spotting device.
The value you provide to me does not always have to be so directly informative or transactional. In fact, in my book, the bar for what counts as “valuable” is generally pretty low. To give an incomplete list of examples, it may come in the form of:
- A lesson.
- A funny story.
- A personal opinion.
- A thoughtful perspective.
- A concern about the present.
- A prediction about the future.
- A piece of information that lets me know you better.
- A question, which lets me fill (or at least, understand) your knowledge gaps, and even often my own.
- A pleasantry, showing me that you may not have anything to say to me, and you may not even care about anything I might have to say to you, but you still respect me enough as a person to offer me a polite sentence or two before dashing off.
On the whole, this system works alright. At the very least, it has, for better or for worse, brought us this far. But it’s not perfect.
One of the ways it breaks down is when two parties have different ideas of what’s valuable. For example:
You yell “cloud!” in my direction.
I listen.
In return, I learn of a cloud. This is not something I care very much about.
In the future, if this sort of thing is a frequent occurrence, then I will be less inclined to listen when you try to speak to me. This is because instead of thinking of you as my friend, I now think of you as a source of loud, distracting, unuseful noise.
Again, this is an exaggeration. We have lots of lovely, polite conventions to avoid this sort of interpersonal disdain: If you do not make good on your end of the bargain and instead offer me some unuseful noise, but I can tell that you are trying to offer me something of value, I will thank you with some courtesies (smiling and nodding, steering the conversation to something more interesting for both of us, and continuing to listen to what you have to say in the future).
Don’t be a meat proxy
The problem comes when I can tell you did not make an effort to offer me something of value. When you spat words in my direction that I now have to process, sift through, and feel obligated to return courtesies towards.
In the present day, for obvious reasons, it is easier than ever to engage in entire conversations without offering anything of value to anyone. This is what the term Meat Proxy refers to: someone who adds nothing to a conversation except latency.
I am sick of this.
If you don’t know how to answer a question, or even how to begin asking it, tell me. Understanding your confusion will always be more valuable to me than words you have no understanding of. If you are baffled, don’t baffle me too.
If you do choose to send me slop, label it is as such, so that I can treat it with the appropriate amount of care. I promise I will not think any less of you, and I will do my best to help you understand them. I will even thank you.
If you do not label it as such, then you have become a strange sort of zombie which is becoming increasingly common: you’re speaking to me (maybe even in person, making eye contact and moving your mouth and so on), and are seemingly processing what I am saying, but actually you are a lifeless computer in disguise, and sooner or later, I will find out.
And then I will prompt you. I will skim what you send me, at most. I will go out of my way to avoid speaking to you, and if I must speak to you, it will be in short, spiteful sentences, with only as little care as I suspect you put into your communications with me.
The whole thing will be a strange rigmarole which is upsetting for us both, where we spend days or weeks offering each other nothing of value, growing increasingly frustrated with one another, learning and accomplishing nothing at all.
It’ll be a real shame, actually. We could have been friends.
How not to be a meat proxy
I’m a relatively young engineer, but I started programming around 12 years old, and did not use any AI for programming at all until roughly the last 1 year, well after college and most of my most challenging programming projects.
I have probably spent a lot less time programming with AI than I have without it. I have also produced vastly more lines of code since adopting it, and I learned vastly less.
The number of programmers who learned how to program before AI is now a monotonically decreasing number, which is a sad, troubling fact. There’s no joy in the world like creating something yourself, or getting something broken to work using just your brain. There are few things more rewarding than mastering your craft.
That’s a joy that has become less and less frequent for many of us, but it’s important for young engineers to strike a balance between getting things done, learning new tools, and honing fundamentals.
I was recently on a project with dozens of engineers working 10+ hour days, costing millions of dollars in salary a month, in which not a single person properly knew their tools or even really knew what they did. They all interacted with them through their various text boxes, getting summaries of summaries of outputs.
When I joined the project and asked how to use these tools myself, it was days before I found someone who knew. And when I did, I discovered that the tools simply did not work (at least, they did not do what everyone thought they did). Something similar happened twice more in the couple weeks I was on that project.
The fact is that being productive used to come with the side-effect of building skills. Increasingly, it doesn’t. Some are going all-in on a bearish bet against learning. I strongly, strongly advise against this.
Learn how to do all your daily tasks by hand, even if you only ever do it once. Programming is all about abstraction, but anyone who tells you implementation doesn’t matter is simply misguided.
AI increasingly shifts the work needed to implement a change away from the planning phase and towards the testing & review phase. I urge you to be intentional about doing the opposite. It can only pay off.
If skills are so useless, then why are startups and tech companies (including my own – shameless, I know) still on the hunt for great engineers?
My good friend and former colleague Claire Yang has recently written an excellent, thoughtful blogpost about this: AI Reflections of a New Grad Engineer. Notice how she is thoughtful, and she is my friend.
If you are not thoughtful, you simply will not my friend. I hope it doesn’t come to that.