"One of the early victories for me was persuading a young boy of 14 or 15 years of age, named Aaron Swartz, to become the technical architect of the Creative Commons. It took a little persuading, but I told him that this is what he had to do." --Lawrence Lessig
Doctorow, Kottke and others didn't seem to think it was odd that a 14 year old was being flown around the country without his parents present. They pre-gamed at a house instead of a bar because Aaron was too young; the host, then nearly twice his age, later dated him. Didn't end well. And then things really didn't end well.
Epstein made his first contribution to MIT that same year (2002). Lessig arranged for Joi Ito to meet Aaron a year later. According to MIT, Joi Ito met Epstein for the first time one month after Aaron died in 2013.
In the most charitable reading of the situation, Aaron was robbed of his childhood.
> people want fuzzy analog machines with digital controls, it's impossible
people would love it if LLMs were deterministic and never hallucinated. It's just that the technology to do so isn't possible, so we make do with fuzzy analog machines with digital controls because we don't have digital machines with digital controls.
We absolutely have digital machines with digital controls, they're called computers, and the controls are called programming languages etc.
The reason people prefer llms over programming is exactly because it lets them specify their program fuzzily, i.e. it lets them avoid going to the trouble of specifying enough detail to make it deterministic.
We have to make the distinction between specification and execution. LLM programming is great because I can underspecify and the machine will infer the rest. Usually reasonably. People want fuzzy specification but deterministic outputs
i feel like Simon omitted an important part of how Qwen's "reasoning" levels work. they are just one sentence additions/omissions to the system prompt
xhigh -> "Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer."
medium -> no mention of effort (sentence omitted)
low -> "Reasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration."
in my testing this doesn't seem to produce exactly deterministic thinking levels, because it's just a system prompt nudge. i had instances where medium thought longer than xhigh
The models are post-trained on these prompt additions so they’re more structural than thinking of them as “system prompts” suggests. (All LLMs ever see is tokens going in, so even the concept of a system prompt is just formatting they’ve seen in post-training.)
You can also apply fixed token budgets for the reasoning blocks, though it will decrease quality in some cases.
Why not invent a few magic token values for reasoning level instead? It would be like 4 out of a vocabulary of 200k and save like 30 tokens in every prompt
yes of course, I understand that. but I feel like it would've been nice to include in the article because the main point of it is the effort and overthinking
same tbh but i feel like we're still required, maybe not in the same numbers as before, and our job description has just changed. it feels more akin to an "AI Agent Operator" or something similar.
i'm still pretty confident someone like my mom wouldn't be able to do my job even with the same access to all the latest LLMs, so we're still providing some value, just in a very different way. whether the market will reprice the cost of our labor, we will see
If software development were getting easier, first thing I'd expect to see would be a strong downward trend in salaries. Why pay a high mid+ dev rate when a junior+LLM can do the same?
That's yet to happen. 90% of software dev skills are still relevant - AI is, for now, just a productivity boost.
We have been heavily using agentic coding in my place, and what is emerging is people who used to be tech leads or architects are now back to producing code, and what would require 5-10 people now starts as a 1-2 person project. Some of those have already gone live, successfully, so there will be more of that.
Between juniors with LLMs and seniors who don't like AI, I'd hire the junior, but I think the trend in 2026 will be to hire senior people who enjoy producing code with LLMs.
Producing code is an interesting word because obviously they're not typing code anymore. That's where I've landed.
I'm somewhere between 10-60x more productive than I used to be. My day cycles between 3-6 different projects, bug fixes, development cycles. Each one them going ten times faster than I could on my own.
We don't even consider bringing on vendors or saas anymore. We only look at our existing vendors and think about getting rid of them.
There is a rift though. Some teams are still just "playing with it". Our offshore workers who haven't seemed to increase their velocity at all. In the next couple of years there's going to be a pretty big wrecking. Those that are agentic and those who are not.
I always ask myself if I'm still bringing value. But then I look at what I'm talking to the computer about. And the amount of different tech stacks algorithms. Tricky business logic isn't something that you can just pick somebody off the street and do.
For now there is value. Only today Opus 5 suggested a really bad but totally workable approach to an edge problem. It would work by putting a lot of undue stress on infra. People would add infra. A simple pushback reduced a O(n) to a O(1) solution - and that was my value added. I bump into those things regularly.
I think the labs want to address the clueless dev market - hence the "proactivity" in Fable. But it will be a difficult thing to balance.
reply