Programmers use this one trick to make themselves unemployable
There was an excellent post on the Databricks blog yesterday. I've never heard of the company, but they apparently spend a lot of money on AI coding, and like many companies, those costs spiraled out of control. It was obvious from the beginning that this was going to happen. Employees using agentic coding (or slop cannons, whichever term you prefer) were constantly talking about how powerful it was, how much easier it made their jobs, and how many more lines of code they were able to write. The one thing they never talked about was the business value, which I'll define here as the additional profit agentic coding is bringing to the business relative to the best alternative approach.
I knew an article like that was sure to bring out some, shall we say, lower tier comments about how the article is wrong. That's where the title of this post comes from. Anyone that is employed in a role that's mostly programming is putting their career at risk if they've bought into the tokenmaxxing marketing. The top comment on the thread is a good example. (I'm omitting the claim about the cost to the company, which is irrelevant, but as others noted, is way too low given current prices.)
I work at a small startup with essentially unlimited AI spend budget - the entire point is that I should be turning to it at every opportunity since our human labor is so expensive relative to tokens.
I produce the output of 3 or 4 2022 engineers and probably at better quality. So it's easily worth it. Would I save money by switching to GLM 5.2 and such...perhaps? IDK.
I can see how someone would fall into this trap:
- This is the new technology which means more of it is better. Anyone resisting is denying reality.
- This is a new skill that employers are going to want in the future.
- Lines of code are dumped into files at a rate nobody had ever seen five years ago.
- It generally seems to produce code that does the job.
The critical error, and the one that turns into a career risk, is "it's easily worth it". That's based on comparison with "output of 3 or 4 2022 engineers" and that's a deeply flawed metric. Imagine someone saying you should rent because the average price of a house in the US in 2026 is 60% higher than it was in 2020. To state the obvious, buy vs rent should compare the cost of buying and renting in 2026, not the cost of buying in 2026 vs buying at the start of 2020. The business value of a particular approach to AI coding needs to compare with the alternative that would be used in 2026, not with vague guesses of productivity in 2022.
An alternative approach to software development with AI is code completion. The programmer uses their brain to think through the big picture, and then an LLM - not Fable, but one that's much cheaper - goes through and fills in the gaps. That's where most of the work is in programming. No need to spend expensive programmer time figuring out how to get an expensive LLM to kind of do what you want (however much more fun that might be than writing code by hand).
Another alternative approach is to figure out the functionality you'll need and then tell an LLM (again, doesn't need to be Fable) to write a library of small, pure functions. Then the programmer writes the code that uses the library, potentially even using code completion to fill the gaps.
The fully agentic approach is a token furnace once your codebase reaches any scale. The first hundred lines aren't bad. Then at a thousand, you say "boy, those tokens are going quickly". Then at 10,000 lines, the token meter is rising like a Geiger counter at a nuclear waste dump. And 10,000 lines isn't exactly a large code base.
But the writing is not the only, or even the biggest, expense. In the future, that code is going to need to be maintained. Bugs will have to be fixed. Current functionality will have to change. New functionality will be added. It will all be done when the codebase is at its largest. Compare the value of a cutting-edge, fully agentic programmer against a legacy AI code completion programmer that understands the codebase and businesses don't want the agentic programmer anywhere near their code.
There's still room for the legacy agentic programmer at this time. That doesn't mean they should be positive about their career prospects. Rather than sticking their heads in the sand and claiming the article is wrong, they should be reevaluating their workflow, and they should be prepared for even greater movement away from fully agentic coding. Even Microsoft, a seller of AI tokens, is imposing token spending caps. In that world, don't expect comparisons with hypothetical 2022 output to save your job.
Now, you might survive if you have a decade of pre-AI experience as a programmer. For the new generation, the ones being taught that the fully agentic approach is the only way to do real programming in 2026, the future could be brutal.