> ## Content Index
> Fetch the complete content index at: https://pogs.ee/llms.txt
> Use this file to discover other available public pages before exploring further.

# So you're worried about AI?
- URL: https://pogs.ee/so-youre-worried-about-ai/
- Published: 2026-11-22T16:00:00.000Z
- Updated: 2026-11-22T16:10:15.000Z
- Description: The industry is doing real harm that we need to be aware of. But AI also can also change lives for the better, and for me it's a really interesting subject when you get past the surface.
- Author: Laurence Chaffey
- Tags: AI, ADHD, LLMs, OKF, Opinion, News

### I get it. Really.

I see a huge variety of views online. Developers who hate it, saying they have lost the passion for their job as they are squeezed into outputting more. Developers who love it and say they're having more fun than ever. People who say it's helped them. People who say they can't find a job because of it. These are all valid experiences and opinions.

For me what is difficult to cope with is how as usual, unregulated and unpunished capitalism is doing its thing; pushing forward towards progress with no care for what we lose along the way. To talk about all the horrible things these companies are doing it would need its own post, but to mention a couple of things that some people may not be aware of:

- We've seen post-training [affecting the mental health of underpaid workers](https://www.theguardian.com/technology/2023/aug/02/ai-chatbot-training-human-toll-content-moderator-meta-openai?ref=pogs.ee) in Kenya, where they are tasked with reviewing texts, and some images, many depicting graphic scenes of violence and abuse.
- Something else I recently heard about is the joint US-Philippines plan to build the "Pax Silica" facility. This project [threatens to displace hundreds of Indigenous Aeta people](https://www.hrw.org/news/2026/11/16/philippines-us-tech-hub-to-displace-hundreds?ref=pogs.ee) and local farmers from the lands they have inhabited for generations.

![Pax Silica: a planned 1,618-hectare US–Philippines semiconductor and AI hub in New Clark City, and the displacement and land-rights concerns around it.](https://pogs.ee/content/images/2026/11/pax-silica-dark.png) × 

Does this sound familiar? I have to agree with the messaging of [Ghost in the Machine](https://www.imdb.com/title/tt39150080/?ref=pogs.ee) that these are examples of **AI colonialism**, despite necessitating collaboration with the country's own government.

The reality is it's difficult to untangle your life from these harms, AI or not. These are systemic problems that require systemic regulation, and as long as money drives politics, sadly these things are unlikely to improve in the short term.

Some things that are within your power to do:

- **Raise awareness.** Contact your local politicians to raise your concerns. Demonstrate. Talk and write about the issues that you see.
- **Cancel your AI subscriptions.** Avoid funding companies directly that you do not want to support. Using smaller open-weight models with a router like [Cortecs](https://cortecs.ai/models/deepseek-v4-1-flash-1017?ref=pogs.ee) can actually save you money too. There are plenty of chatbot and agent apps on all platforms where you can bring your own API key, which can also have privacy benefits. If you have an efficient computer, you can also try running local models. I strongly believe local models are the future of daily AI use as returns diminish on LLM training.
- **Use more efficient models.** If you haven't already used it, DeepSeek V4.1 Flash is very fast and suitable for complex work. Data from [Know Your Compute](https://www.knowyourcompute.com/compare/?ids=claude-opus-5.5,deepseek-v4.1-flash&ref=pogs.ee) suggests that it uses approximately 4.5x less energy, 2.2x less water, and outputs 3.2x less CO2 than Opus 5.5\. Video generation models especially are some of the worst contributors. Another recent development is Decision models, which can be harnessed by software to do a lot of what a normal LLM can do at the fraction of the cost and size. I'll be writing a post about these later.
- **Use AI less.** There are a lot of great things that LLMs can help us do, and you can't put the genie back in the bottle. But not everything in your life has to be automated and summarised. Keep doing what you love, and just let it do the things you hate. When used correctly, as one of my favourite writers [Lawrence Yeo](https://moretothat.com/?ref=pogs.ee) explains, AI can be [an extension of your creative self](https://moretothat.com/ai-is-an-extension-of-your-creative-self/?ref=pogs.ee).
- **Teach your agent to fish 🎣.** No, not literally... but the old saying goes here. If your objective can be accomplished with software, and you don't know how to code, then get your agent to build a tool that accomplishes the task at least partially, instead of repeatedly churning tokens on something that code can do. An example of this would be using a script to scrape and sanitise data before parsing it with an LLM, instead of having the LLM do the whole thing, or using an RSS feed to gather articles that you want to summarise instead of trawling the web itself.

---

### The positives.

I want to move past the doom and gloom and talk a bit more about my own experience with AI. There's two things in the last couple of years that seriously changed my life. First was a late ADHD diagnosis and getting medicated. Second is, you guessed it: **AI**.

I'm a technical person at heart, with a background in IT support, system administration, and now cloud solutions. But I never really got the hang of coding. I can read and understand it but I was never able to write it independently. I still struggle to concentrate on studying. I've picked up most of my skills through experience and hard work. So when AI came along and addressed one of my key weaknesses, I was pretty chuffed about it. But it does a whole lot more than coding for me.

There's a lot of talk about AI making people dumber, and I've no doubt it's true if you're lazy with it. If you offload all of your thinking on to an LLM, yes, you'll probably get worse at thinking. But for me, AI has pushed me into doing things I'm not sure I would be capable of without it. Both creative *and* logical endeavours.

One of my biggest problems is that I'm a serial perfectionist. I find it difficult to set out on a task if I haven't planned out every detail. I'll spend a long time thinking about the the big idea, and before I know it something else pulls me away before I even start.

One thing that AI has been amazing for is exploring pathways. When I have an idea, I can now talk it out, ask about the strengths and weaknesses of certain methods, and get the information out of my head. This frees up a lot of my brain cycles.

The killer feature for me though is *examples*. Being able to generate an example of how a system is used, or where a piece of code fits, has been the single most valuable thing for me. Years ago I would have had to bother a senior developer to push my knowledge further or hope I could find what I needed on Google. It's like having your own personal tutor.

I found the last two posts from [Terrible Software](https://terriblesoftware.org/?ref=pogs.ee) summarised quite well how I feel about using AI to learn and work.

### First - [The Brain Sandwich](https://terriblesoftware.org/2026/10/02/the-brain-sandwich/?ref=pogs.ee). 

As the author Matheus Lima writes: 

> whenever you work on something, you should use:   
> 1\. Your brain, *and then*  
> 2\. AI, *and then*  
> 3\. Your brain.

It sounds really simple, because it is. I've been using AI a lot to dive into new concepts, and I learn so much by thinking and probing the output with more questions. If you are skipping step 1 or 3, you are doing it wrong.

![The Brain Sandwich concept.](https://pogs.ee/content/images/2026/11/brain-sandwich-dark.png) × 

### Second - [Why Am I So Tired If AI Is Doing All The Work](https://terriblesoftware.org/2026/11/11/why-am-i-so-tired-if-ai-is-doing-all-the-work/?ref=pogs.ee)?

As part of my job, I am doing a lot of work building efficient AI agents to help us scale. Even though my coding agent was doing most of the work, I still felt like I was context switching a lot. Watching it, Slack pings, emails, and whatever else was wearing me out. I had placed an expectation on myself that I could do my normal day-to-day work and not worry about doing more. But as Matheus explains, that's just not how it works:

> I think we’re making this worse by saying yes to things we would’ve simply put on our priority queue (or just left alone) before. Because “I’ll just ask my agent” feels like such a small commitment — and easier than creating a ticket, right? — that **we forget we’re also volunteering to understand, check, *and* finish whatever comes back**.

So the important lesson here is: don't use AI to do to a more work. It's still work, it's still context switching, and it'll still burn you out if you're not careful. Keep focusing on the one or two things you value most. Use AI to *improve* your output, not to increase it. If you choose not to use AI at all, I'm not going judge, but I truly think everyone has something to gain from it.

---

### Closing note

Linking back to my earlier point about efficient models, you may have noticed the chatbot at the top of this website. Well on my two CPU cores, I am running [Bonsai 2 4B](https://prismml.com/news/bonsai-2-family?ref=pogs.ee), a highly compressed version of Qwen3.5-4B that makes use of the spare cycles this website is running on - hence the name. PrismML's Bonsai models are outstanding for their size, able to run comfortably on CPUs and even mobile devices. We'll have to see how well it scales...

It's backed by an agent harness that uses local knowledge indexed using Google's [Open Knowledge Format](https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing?ref=pogs.ee) with a retrieval tool. It verifies answers to catch out hallucinations, and caches them so it never has to answer the same question twice. It doesn't have a lot of data yet other than what you would find on my resume, but I'm working on expanding it. I've loved using OKF as it's easy for both LLMs *and* humans to maintain.

Over the following weeks I plan to go more in depth on the subjects discussed here. This will include breakdowns on building efficient agents and deep-dives into LLMs, as well as the impact of technology on society and our psychology.