Now might be the best time to learn software development

And the most fun

What do you do while awaiting the agents writing your code?

ChadGPT suggests exercise, and so I drop and do twenty. Planks are good too but tempt me to use Gemini Flash where a thinking model would be better. And it's hard to explain at the company summit why I'm suddenly sporting washboard abs and biceps the size of hams. "Oh, uhh, definitely NOT AI," I lie, blushing profusely, "The only weights I use are at the gym! Haha."

How many coding agents can you manage simultaneously?

I find it hard to manage more than three before I start replying "fix it!!" without really understanding what went wrong. While they work I’m typically reviewing code (and reverting half), writing new rules (“plz grep for components before writing them”), or checking out new agents (platonically).

I’m having a blast; am I doing something wrong? Aren’t developers supposed to be miserable? I suppose there are storm clouds yonder; AI wars and rumors of AI wars. I’m told the very models I’m having fun with today will be the reason I’m laid off tomorrow, that some execs are jumping the gun and pulling the plug on new talent already. Evidently I've been a highly paid craft farmer my whole life, and the LLM is the mighty combine harvester, here to harvest my codes. Or plant them and then harvest idk what an analogy is.

Tom says it best:

Apprentice a tradesman or die. Down with the rich farmers! Just like that period of history where farmers were way overpaid. Great metaphors always build on such uncompromising premise.

LLMs really are like combine harvesters; allowing one to do the work of many. And like mechanized farm equipment, LLMs are cheap, plentiful, getting smaller every day, and- most importantly- require no training to operate. "Hey CombineGPT, teach me to reap the vast fertile plains!" And off you go.

“Okay snarky man, if LLMs are so easy and plentiful why keep devs at all hmm??” Fair point. I am just a middle-man, the hired hand here to corral the cattle on Software Ranch until the day they corral themselves (it's a cowboy analogy now ig). "Go become fine steaks and leathers!!" one boss in a stetson will yell, and the cows will oblige. Devs will take the lonely road into the sunset, and historical romance novels will rightly remember us as rugged and sexy.

I would resign myself to the role of romance-novel-hunk but I think there’s a strong case to be made that the death of the software developer has been greatly exaggerated. In fact, I suspect the opposite is true: today might be the best time in history to learn software development. Let’s support this bold claim with commensurate bold points:

1. It’s your moat, too

This may end up as some obvious hindsight. "You know how our developers have all this institutional knowledge and intimate understanding of our business and know how to build our products better than anyone else? Well they're expensive and it turns out our product is easily written by a cheap-as-free AI service. Yeah the devs do know how to wield that AI better than anyone but they're expensive so let's fire most of ‘em and rejoice in the great savings as our business continues undisrupted forever!!"

The moat protecting your devs is also protecting your business. If AI has removed your moat then you can be sure disruptors are approaching your gate. Is now really the time to start firing your soldiers? Your sexy cowboy soldiers??

Actually, it's more accurate to say that the moat is still there- it's just that crossing moats is now trivial since every soldier suddenly has an attack helicopter (it doubles as a combine harvester please keep up). Devs are gonna develop- and now with far greater reach and speed than ever before. So my advice to any exec thinking about layoffs: you can be sure that other institutions are reorganizing their AI-powered devs to conquer new territory, why aren't you doing the same with yours? Are you really gonna play defense when everyone’s getting nukes?? (Nuclear attack combines now).

My advice to devs is the same: Don’t wait around to be laid off by a bot. Pick up AI like some kinda chosen one drawing your sword from a stone- you’re definitely the person most capable of wielding this shit. This is your time! Yeah change is scary, but you’re a technologist. Or did you sign up thinking tech would always stay the same?

2. AI grants wishes, developers discover

Building an app with zero dev experience? Use AI and you will succeed. You will: the app will be built. AI will crank out fully functional code and you might even feel a tingle of power. But here’s the catch: writing code is the easiest part of a developer's day.

The vast majority of the problems a developer solves are social. This is because nobody sees reality correctly yet everyone insists they know what they want. We’re all groping around in the dark and LLMs create more illusions than illumination.

A medical provider asks AI to make a better record system. Voila- here’s a blockchain based app with 2048-bit security and convenient forms for every case! There is no migration strategy or auth integration, and all the staff still share passwords for email accounts with no 2FA. But the app feels right.

A contractor wants a portal complete with real-time updates on all client projects. Fantastic, AI chooses Phoenix for its lightning fast realtime capabilities, complete with a powerful ecto data migration toolkit. All the client data still lives in an Excel file that you update when you get around to it, but the portal is great. Next month a zero-day is published for one of the deployment packages, but you never hear about it; who would even ask?

You tell an AI your application needs to be GDPR compliant- you aren’t really sure what that means, but the AI does- and so it dutifully constructs a workflow for data deletion. Fantastic! Now customers can click the Delete button and feel like their data is no longer retained. The next day you have another idea: “Let’s backup customer records to my home server, I don’t trust Amazon and I think we need to be in control of our own data!!” The AI complies, throwing in some effusive praise for your strategic thinking.

These examples might feel too silly and contrived to be real, but I promise that developers have to talk clients down from this kind of bullshit every single day. Devs spend far more time discovering reality underlying the problem than writing code to solve it. And this is good: better to spend 5 days finding the right problem than 1 day solving the wrong one.

LLMs aggravate these misdirections. They sometimes strike underlying reality, but most often they just create The Thing That Comforts You Most. It takes deep understanding and experience to know when something is actually helpful or just the illusion of a solution, and even devs can be fooled by those long blocks of clean code and shiny reactive buttons.

"But AI will just teach me what I need to know!" - Yes! And with a lot of careful trial and error you, too, will become a developer. AI can certainly make learning easier, but the more you neglect learning foundations the longer you will spend wandering The Land of Large Language Illusions.

3. Software is kinda the last problem anyway

This is a little more grim but really- we still have a hell of a lot of software problems to solve, and once they're all solved we probably won't have any job that AI can't do better. Lumberjacks are going to be replaced by software, not unemployed software developers.

AI has so much amplifying ability for the person who knows what to ask. There are so many great new tools, and so many terrible ones that sound great on paper. Can you spot the difference? Even the good tools can be extremely frustrating when they always seem so close to giving you what you want but are never quite there. It's a fantastically fun time to build things, for those with the patience to do it well.

Learning has never been easier- you can now consult an oracle that knows every language and has memorized all of StackOverflow. So it bothers me that all we can talk about is how grim the future is for software developers, discouraging students and shuttering talent pipelines, when such talk might be killing off talent precisely at the moment when it is easiest to learn, most productive to use, and more important than ever to have humans holding the reins.

And why? Because of speculation: that development practice will die instead of advance. That there’s some zero-sum pie of problems worth solving. That with just a little more progress all of the problems endemic to AI will suddenly vanish.

Massive amounts of capital are shifting into AI tech. Neat! But it’s a mistake to stop investing in traditional devs, and I think the organizations that keep investing in talent will ultimately win. I think the first generation of AI-native programmers are gonna be some arcane wizards by the time they're in their thirties. And if no AI comes riding in to rescue those that don’t invest in new talent... well, then we will see what a truly overpaid farmer costs.

It’s still a gamble: the only thing sure about technology is that it’s changing, and anyone who says they know where it’s going is just guessing. But even if humans-in-the-loop stop adding value someday, I remain convicted of their importance. This applies to all professionals whom AI is supposed to replace: it is not enough to use AI, we must have experts who can hold it accountable, who can recognize when it is giving comfort instead of truth. The day we stop valuing human contribution is the day alignment has failed.

I’m obviously a man of few words, so I’m gonna stop there. Time for this cowboy to climb into a nuclear attack combine and ride off into the sunset. Or is that a sunrise? It could be a sunrise.