Software development in the time of strange new angels

Completely agree on your first point: software development is so much more than writing code. LLMs are a threat to programmers for whom the job is 8 hours a day of writing code to detailed specifications provided by other people. I can't remember any point in my own career where I worked with people who got to do that.

There's a great example of that in the linked post itself:

> Let's build a property-based testing suite. It should create Java classes at random using the entire range of available Java features. These random classes should be checked to see whether they produce valid parse trees, satisfying a variety of invariants.

Knowing what that means is worth $150/hour even if you don't type a single line of code to implement it yourself!

And to be fair, the author makes that point themselves later on:

> Agentic AI means that anything you know to code can be coded very rapidly. Read that sentence carefully. If you know just what code needs to be created to solve an issue you want, the angels will grant you that code at the cost of a prompt or two. The trouble comes in that most people don't know what code needs to be created to solve their problem, for any but the most trivial problems.

On your second point: I wouldn't recommend betting against costs continuing to fall. The cost reduction trend has been reliable over the past three years.

In 2022 the best available models was GPT-3 text-davinci-003 at $60/million input tokens.

GPT-5 today is $1.25/million input tokens - 48x cheaper for a massively more capable model.

... and we already know it can be even cheaper. Kimi K2 came out two weeks ago benchmarking close to (possibly even above) GPT-5 and can be run at an even lower cost.

I'm willing to bet there are still significantly more optimizations to be discovered, and prices will continue to drop - at least on a per-token basis.

We're beginning to find more expensive ways to use the models though. Coding Agents like Claude Code and Codex CLI can churn through tokens.

Both the article's examples there are bogus -- yet in both cases the underlying points are true.

Google generates a lot of revenue per employee not because the employees are good (though many of them are of course), but because they own the front door to the web. And the Knight Capital story has many nuances left out by that summary.

In both cases the author needed a hard hitting but terse example. But as I said, both the claims are true, so in the voice of the courtroom judge, "I'll allow it."

This. AI is not replacing us, it is pulling the ladder up behind us.

> Having something that can generate code reliabily (which these things can barely do even with an expert at the wheel) doesn't address any of the actual hard problems we deal with on a daily basis.

Not understanding that is something I've been seeing management repeatedly doing for decades.

This article reads like all the things I discovered and the mistakes the company I worked for made learning how to outsource software development back in the late 90s and early 2000s. The only difference is this is using AI to generate the code instead of lower paid developers from developing nations. And, just like software outsourcing as an industry created practices and working styles to maximise profit to outsourcing companies, anyone who builds their business relying on OpenAI/Anthropic/Google/Meta/whoever - is going to need to address the risk of their chosen AI tool vendor ramping up the costs of using the tools to extract all; the value of the apparent cost savings.

This bit matches exactly with my experience:

"The trouble comes in that most people don't know what code needs to be created to solve their problem, for any but the most trivial problems. Who does know what code would be needed to solve complex problems? Currently that's only known by software developers, development managers and product managers, three job classifications that are going to be merging rapidly."

We found that assuming the people you employ as "developers" weren't actually also doing the dev management and product management roles was wrong. At least for our business where there were 6 or 8 devs who all understood the business goals and existing codebase and technology. When we eventually;y got successful outsourced development working was after we realised that writing code from lists of tasks/requirements was way less than 50% of what our in-house development team had been doing for years. We ended up saving a lot of money on that 30 or 40% of the work, but 60 or 70% of the higher level _understanding the business and tech stack_ work still needed to be done by people who understood the whole business and had a vested interest in the business succeeding.

> In 2022 the best available models was GPT-3 text-davinci-003 at $60/million input tokens.

>GPT-5 today is $1.25/million input tokens - 48x cheaper for a massively more capable model.

Yes - but.

GPT-5 and all the other modern "reasoning models" and tools burn through way more tokens to answer the same prompts.

As you said:

> We're beginning to find more expensive ways to use the models though. Coding Agents like Claude Code and Codex CLI can churn through tokens.

Right now, it feels that "frontier models" costs to use are staying the same as they've been for the entire ~5 year history of the current LLM/AI industry. But older models these days are comparably effectively free.

I'm wondering when/if there'll be a asymptotic flattening, where new frontier models are insignificantly better that older ones, and running some model off Huggingface on a reasonably specced up Mac Mini or gaming PC will provide AI coding assistance at basically electricity and hardware depreciation prices?