There’s a lot of angst around AI at the moment. For us programmers, AI has gone from a helpful “idiot savant” to a geniune self sufficient programmer. Then on Oct 6th, OpenAI published ~700 math papers generated by a model not yet available to the general public.

It won’t take much digging to find negative sentiments about this. Grant Sanderson of the youtube math channel 3Blue1Brown released a positive “There’s never been a better time to go into math” video. However, scroll down a bit and read an emotional counter-point:

@alvih333

I feel like the point is still being missed […]. The argument you are making “Well yes now this part of math we will not need to do again, but now we can do this instead” however the issue with most people just don’t want to do that…

Like who are we kidding, the type of nerds who were attracted to math and software were into it because we saw it as an art-form, it was a craft, it was being the only one in your city to be able to solve a complicated math problem, it was immortalizing yourself because you discovered a proof and added to the ever expanding human knowledge.

[…]

A craft I enjoy doing has been taken away from me, I should be allowed to mourn, I don’t care that I can do something else since I never liked doing it in the first place. I enjoyed solving problems and writing code, I don’t care if people say “well thats never what programming was about it was actually about becoming a manager all along” I never wanted to be a manager, I never wanted to manage, I never wanted to be a proof reader.

I just hate this existence.

I don’t agree with the comment’s closing sentiments. This person can still go and enjoy doing mathematics as a hobby. However, I agree with their sentiment if they’re lamenting that they won’t be able to enjoy this craft as a job.

I believe in the next 10 years we are going to see a lot of programming jobs disappear. I don’t know if they will be replaced partly or fully with other jobs, like the managers or sales roles the comment alludes to. I’m guessing there will be a lot of new work similar to todays Quality Assurance engineers.

If that’s surprising to you, let me convince you of the essence of a programmer’s job.

It was always about the programming and it was never about the programming

For the past 50 (or more) years we’ve had a status-quo: to make an executable program you HAD to go employ a programmer.

Then, in a swift time-span of 2 years the AI models have gone from

1. A semi-helpful chatbot in a separate app, where the programmer could copy and paste text to and from.
2. A claude/codex program that can read a project’s code or decode inscrutable error messages. It was then living in the programmer’s code editor.
3. A bona fide agent that can be given a high-level goal and it will very reliably produce a working solution. Nowadays most programmers don’t open a code editor app.

The crux of the employability issue is in the working solution.

It completely changes the status-quo: you do not need a programmer anymore.

down

B: [ 
  box "business" 
  Employs: line ->
  box "programmer" fit
  Produces: line -> 
  box "working program" fit
  text at 1mm right of Employs "employs" ljust
  text at 1mm right of Produces "produces" ljust
]

text at 1cm above B.n "The status quo" big "for the past 50 years" big

dot invisible at 5cm right of B.n

A: [
   box "business" 
  Employs: line ->
  circle "AI" fit
  Produces: line -> 
  box "working program" fit
  text at 1mm left of Employs "pays" rjust
  text at 1mm right of Employs "API" ljust
  text at 1mm right of Produces "produces" ljust
]

text at 1cm above A.n "Today" big

From a manager’s perspective the programmer’s job in the second phase was clear: the baby sitter. The manager understood that claude/codex was doing most of the “work”. They weren’t even mad about it. They were able to close a job posting for the junior role since the baby sitter and claude/codex took care of the unglamorous work that’s usually ignored and instead shoved onto the junior as “learning opportunities”: fixing small design discrepancies, chasing down rare issues that were hard to reproduce, updating dependencies, etc.

In the 3rd phase, I’m not even sitting on that baby. It just prints and prints and prints and prints working programs faster than I ever could.

The thought has probably crossed the manager’s mind more than once a day. The programmer is not needed anymore. That same cost-trimming-budget-optimization axe that struck the junior’s job posting last week was used on the “Senior Software Engineer” job posting today. I’m not great at math but even I can count 1-2-3.

To be clear, this didn’t happen to me but it’s a scenario that’s playing out throughout the industry.

From a business perspective, it seems like a no-brainer. It’s MUCH cheaper to have the AI produce the code. After all, it. just. works.

The gotcha here is of course that programmers do more than just write programs. In much the same way that mathematicians do more than “convert coffee to theorems”.

The essence of the programmer, the baker, the plumber

Much better writers have broached the subject before me. See the blog post “Do You Still Read the Code?” by Marek Chotoborski or the 10-page article from 1984 by Peter Naur Programming as Theory Building.

I will take a more illustrative approach.

I’ve already mentioned that to make an executable program you needed a programmer before. As such, it’s easy to assume that’s their only role. This is a fallacy. The executable program is merely an asset that’s created. It is not what makes a programmer valuable. Their experience, skills, and context are what make the programmer valueable.

Consider other carreers.

Is the baker revered because they made good bread? No. The bread is the asset they created. The value of the baker is their bakery and their process. Today’s bread is made using the dough mixed yesterday. It’s given a day to proof and rise. This is a logistical nightmare. How much dough do we need? A bakery sells not one type of bread but up to a dozen, and other baked goods. Each type of bread takes a different amount of time to bake. Some use different doughs. To meet the rush of customers at different times of the day, what is the order of baking and when do they fire? It’s a marvel to be able to buy fresh bread everyday.

Is a plumber valueable because they laid pipes in new construction as the build plans dictated? No. A plumber is valueable when they can walk into an 80 year old home, diagnose the cause of a leak, the source of the smell of sewage, the low water pressure, etc. It doesn’t matter who built it and whether they used copper pipes, lead pipes, cast iron, plastic, or an unholy amalgamation. The plumber can envision the trade-offs, make all those tiny decisions, and propose a solution in a tidy bill.

Programmers are not unique. They face the same types of problems.

Like the baker, they understand some data has to be pre-processed to be instantly served on demand. A youtube video is uploaded as one file but becomes available to users in many different resolutions, with auto-generated captions in multiple languages.

Like the plumber, they understand that there’s dozens of programming languages, hundreds of frameworks, and thousands of libraries. They see the problem space and craftily choose a set of these to start a project or augment it.

Has your AI ever told you no?

Today, AIs are wonderfully goal-oriented. This science-fiction-level of agency leads people to give them autonomy. This is a mistake.

A key part of autonomy is building your own perspective, your own theory and defending it when it matters. It’s very difficult nowadays to provide enough context to an AI system to be able to do that effectively. I believe it will continue to be a significant problem for at least 10 more years.

We do not like to be told no, but it’s necessary. No beautiful snowflake idea ever survives the gritty reality of the real world. Our intentions are seldom aligned with the different grating compounding factors of our messy interconnected world.

Customer: I’m sick of being cold in the winter and hot in the summers. I want to add insulation to my house. Let’s add outdoor insulation instead of indoor insulation. This way we avoid encroaching on the limited and valueable internal space

AI: Great choice. Tell me the outdoor surface area to cover and what your budget is. I’ll search materials with the highest R-value that matches the budget to maximize the insulation!

Builder: I highly recommend indoor insulation instead. From my inspection, your house has had humidity problems before. Outdoor insulation tends to trap water vapor, specially the materials with higher R-value. All outdoor materials eventually wear out. This leaves gaps and once water seeps in, outdoor insulation essentially forces the humidity inside. The house cannot breath out the humidity. You will have mold problems and the solution will be expensive repairs on the outside or…switching to indoor insulation.

Note: I’m a programmer, I don’t build houses. This is a hypothetical scenario.

The essence of the programmer’s work is to take the business’s goals and requirements, the intended user experience, the performance budget, the time budget, and the budget budget into consideration and make the technological choices. It is to build that context, make the executable programs, and deploy it. Test it live and check assumptions from all these stakeholders. Understand how each part of the program matches to the real world and matches to the design considerations. Then use that understanding to destroy, mend, or create anew when our naive assumptions inevitable crumble under the entropy of the real world.

Programmers are most valuable when they tell their leaders “No”. To surround yourself with sycophantic souless yes-machines is to enable the limitless potential of your lack of experience and intuition.

The programmer is paid not to write programs specifically but to create the “theory” of the program. This theory is abstract, like a thought that we can never put into words quite right. It’s both useless by itself and indispensable to the programmer. The theory guides the creation of the program as it gets closer and closer to the ideal solution as we better understand the problem we are trying to solve.

The programmer should understand the world as it relates to the problem at hand, and point to the part of the code that wrangles with that reality.

The programmer’s value is apparent when our perspective on the world changes. The problem definition updates, the business pivots, and the programmer with a solid theory can start making a list of the changes that are needed.

If you are unconvinced, I highly recommend reading Programming as Theory Building by Peter Naur. It’s on 10 pages long and still incredibly relevant despite the fact it was published in 1984.

So…our jobs are safe?

Are programmers’ jobs safe? Nope. Well…not the nice cushy roles that we are used to.

The last 20 golden-age years of venture capitalist fueled salaries is probably gone for good.

The vast majority of businesses have no clue what they’re doing. They’re adulting as hard as we are. Most need to execute faster, iterate, measure, and pivot. A cheap and quickly executing AI is exactly what they need.

Established businesses will still need programmers. A multi-million dollar company doesn’t need to innovate as much as it needs to keep the lights on, be reliable, and maintain their customers trust. They want an expert to make the right choices along those axes. If they instead chose to cost-cut with AIs driven by middle managers, that’s their grave to dig.