I feel like I am getting burnt out from AI at the moment. It is everywhere in tech and if you work as a software engineer you are no doubt using it or being told to use it every day.
It is not that I fear I am going to be replaced by AI, but more that it is going to replace the parts of the job that I actually enjoy.
Yes part of getting into programming was because I enjoy building things and with AI you can build things a lot faster, but what is lacking is the problem solving, coming up with creative solutions and the craft.
It is no coincidence that the first issue of this newsletter was titled, "Programming = Creativity 🎨". I still believe that programming is a creative profession that involves solving problems and designing solutions.
I definitely don't fit completely into the anti-AI camp, I use AI most days but not every day.
Mostly I use AI for:
- Prototyping some UI concepts.
- Writing boilerplate code such as converting JSON into Typespec of C# classes.
- Writing unit tests, with a fair bit of hand holding.
- Asking about the odd obscure bug.
- Writing one off scripts for converting data.
The main thing is that I don't let AI do the actual work, especially if I am going to the be one responsible for it, and it has my name associated with it.
A lot of these use cases also don't need frontier models and can be run with a local model on your own machine.
I have never enjoyed reviewing code, other than for junior developers where it can be used as a learning exercise to share knowledge. Reading and understanding someone else's code is always harder than code you have written yourself.
At least when the code is written by a trusted developer you would hope they have done some due diligence and actually tested their code before putting for review. With AI written code you have none of these guarantees.
It feels like everyone is obsessed with agentic coding at the moment. Even those who have never written a line of code in their life are vibe coding applications and prototypes that do look from the casual observer quite impressive.
In reality though, there is large gap between a prototype you have put together in Claude Code to making something production ready. It is always the last 10% that takes the most time. For the non-engineers who have vibe coded something in an hour, this nuance is lost. There is an assumption here that coding is easy, if I can do this in an hour why aren't you doing the same?
As always, technological advancements aren't giving us back time they are demanding more out of our working hours to the point of burnout. Even if you don't want to use AI to write all the code for you, the bar has been raised as to what it is expected of you. Code quality and ethics be damned.
At the end of the day it isn't your manager who is going to be on call if the AI written code has a bug in it at 1 am.
❤️ Picks of the Week #
I think this is a large part of the AI burnout. I don't select AI articles on purpose but 29 of the 39 links below are directly related to AI with a few other indirectly related. It is just the nature of the news at the moment.
📝 Article - A new California law says all operating systems need to have age verification - Age gating certain websites I understand to a certain extent, but bundling into the operating system is a big no from me.
📝 Article - Leaving Google has actively improved my life - I pretty much completely de-googled last month. Obviously my YouTube channel is still tied to a Google account, but it is a free one which I don't use for emails.
📝 Article - MCP server that reduces Claude Code context consumption by 98% - I haven't found a need to use MCP servers in my limited use of AI, but if you do, you might want to look at this.
📝 Article - Microgpt - So much of AI is a black box. This is worth a read to understand how these big models work at a fundamental level.
🧩 Demo - I built a demo of what AI chat will look like when it's “free” and ad-supported - At the moment the AI companies are running at a loss. It is like the early days of most services. At some point they will need to introduce ads or up the costs.
📝 Article - AI Made Writing Code Easier. It Made Being an Engineer Harder - I think most senior engineers know that lines of code are a liability. Reviewing code for mistakes is a lot harder than actually writing the code yourself. The expectation of what can be produced just increased as well.
📝 Article - When does MCP make sense vs CLI? - MCPs are essentially simplified defined APIs for LLMs. If they can use CLIs then that is often the better option and apparently cheaper in terms of token use as well.
📝 Article - The nature of the job - This sums up my thoughts exactly. I don't look forward to Monday's any more. I would like to get back to enjoying programming which involves problem solving not babysitting an LLM.
💬 Thought - Fucking Vibe Coding - I would not want to work at one of these companies. Although I can see it is difficult to work at any company that doesn't have similar thoughts.
🛠️ Tool - Code Corners - This is quite cool, especially as I try to avoid hosting projects on GitHub as much as possible.
📝 Article - Ars Technica fires reporter after AI controversy involving fabricated quotes - I mentioned this article a few weeks ago. I am glad Ars Technica took a stand against AI generated content. We really don't need more AI generated articles.
🧩 Demo - The Xkcd thing, now interactive - I love this, it would be great to see more of these.
📝 Article - I'm losing the SEO battle for my own open source project - This is sad and just goes to show how far Google and the like have fallen. If you suffer from this the correct thing to do is to go round and try and email everyone who is linking to the wrong project.
📝 Article - Agentic Engineering Patterns - There is some good advice here and most of it applies even if you aren't doing agentic coding. Things such as use TDD and not inflicting unreviewed code on reviewers.
📝 Article - Nobody Gets Promoted for Simplicity - I have seen this a lot throughout my career. The code that just works doesn't get as much credit as the overengineered solution. I find writing a list of your achievements throughout the year can help a lot.
🎓 Guide - Qwen3.5 Fine-Tuning Guide - When it comes to AI I would much rather use a local model than a cloud one that I have to pay for. There a probably hundreds of use cases where small fine tuned models could be useful.
📝 Article - The L in "LLM" Stands for Lying - If a fake painting is forgery than a LLM produced code is a forgery too. The only reason AI code works is because it is copying from someone else.
📝 Article - A GitHub Issue Title Compromised 4k Developer Machines - Engineering rule number 1. Don't trust user input. Yet here they are getting code generated and committed from user input 🤦♂️.
📝 Article - Glassworm is back: A new wave of invisible Unicode attacks hits repositories - We are likely to see more of these attacks especially as developers struggle to properly review code due to the sheer number of changes due to AI.
📝 Article - Stop Sloppypasta - This is a pet peeve of mine. If you can't be bothered to write it then why should I be bothered to read it.
📝 Article - Being a Luddite Is Cool and All, but Have You Seen the Hilarious Tapestries These New Looms Are Making? - I loved this article. Many seem to justify the uses of AI based on the fun things it can do. The whole thing has been designed by billionaires to get rid of as many workers as possible.
📝 Article - Every layer of review makes you 10x slower - Reviews are definitely the bottleneck and they should be. They are the last line of defence before slop gets committed. The larger the code review the harder it is to review and insure that you haven't missed anything.
📝 Article - Software Harm Reduction - I think it is going to get harder and harder to find software that hasn't been touched by AI. I am not totally against using AI in your codebase but it is a slippery slope between generating boilerplate code and vibe coding whole features.
📝 Article - A sufficiently detailed spec is code - This is the crux of the matter. If you really want an LLM to produce code that works properly then you need to be very specific in your prompt. To the point where you are basically writing pseudocode anyway.
🛠️ Tool - Three new Kitten TTS models – smallest less than 25MB - Text To Speech is one of those AI areas that I think is OK morally speaking as long as you avoid cloning other peoples voices for nefarious purposes....
📝 Article - Anthropic takes legal action against OpenCode - Anthropic don't really have a mote, all the frontier AI models are pretty good. They are desperately trying to lock people into the ecosystem so they can raise the prices later.
📝 Article - Trust and Faith in Our Web - When AI first came out, I used to generate images for my blog posts, but I removed them all and left only those with my own drawings (#1, #2). If I ever make enough money for this I will hire some artists for my newsletter issues. Sacha's work looks particularly good!.
📝 Article - How I'm Productive with Claude Code - I like to see how other developers are using agentic coding tool to see what I am missing out on.
🛠️ Cheat Sheet - Claude Code Cheat Sheet - A useful cheatsheet for those using Claude Code.
📝 Article - LiteLLM Python package compromised by supply-chain attack - I do use LiteLLM, luckily the Docker version isn't compromised.
📝 Article - So where are all the AI apps? - I think a lot of code is getting produced with AI but very little makes it to production. There is a large gap between vibe coding something that works on your machine and getting it ready to be used by others.
🐦 Tweet - Goodbye to Sora - There are still plenty of places to generate AI videos. I have tried local AI videos, and it took half an hour to generate a 5-second video while my PC screamed at me. I dread to think of the environmental impact of these video generators.
📝 Article - Is anybody else bored of talking about AI? - 👋
📝 Article - Thoughts on slowing the fuck down - A large part of an engineers job should be thinking about the problem and coming up with an elegant solution for it. People seem to be delgating their thinking in the name of speed and productivity.
📝 Article - Apple Just Lost Me - I have turned off automatic updates on my iPhone due to this new update. I do have a credit card attached to my Apple Wallet, so hopefully that will be enough.
📝 Article - Quantization from the Ground Up - If you are trying to understand quantization then this is a great article.
📝 Article - 90% of Claude-linked output going to GitHub repos w <2 stars - This is an interesting dashboard showing the increase in AI generated code. Of course most of the code is going to be in new vibe coded projects.
📝 Article - Updates to GitHub Copilot interaction data usage policy - Of course they are using your code to train AI. This is why we should limit ourselves to local models where possible.
📝 Article - The EU still wants to scan your private messages and photos - It sucks that this is the world we live in where privacy is dead.
💬 Quote of the Week #
The rational solution would be as soon as the necessaries and elementary comforts can be provided for all to reduce the hours of labor gradually, allowing a popular vote to decide, at each stage, whether more leisure or more goods were to be preferred.
Circling back to this quote from Bertrand Russell's article “In Praise of Idleness” from 1932. I will happily let AI take over the work if it means we all get UBI and can work fewer hours. I don't think that is the direction this world is going in unfortunately.
Top comments (0)