mark_l_watson

While I agree with the article, I would carry the idea further: the huge value in AI is improving ourselves. By all means use AI to help design, write tests, and implement code and while I understand this might be the big value add for corporations that may employ you, the important investments in life are in learning and self improvement. I built a good career primarily improving my own knowledge and skills, and along the way doing some useful work for my employers. I am an old man and retired so now I 100% use AI for learning.

Thanks, you make me feel better for spending the (fun!) time writing my first harness in Emacs Lisp with a ‘Emacs UI’ and later writing a command line harness in Common Lisp. Am I more productive with my own harnesses? Not yet, but the second harness I wrote in Common Lisp is promising, and I might write a little book just on this project to encourage people to hack what I wrote and make it their own.

Odd results for me. Last month I tested five AI chat sites, with web search turned off, and four of them had a shadow of information about me as a person (what kind of books I write, what tech I use, and a random bit of other information). The linked site gave me a zero score because it was testing if the AI models recommended my site for business or sales queries.

Such great advice. I have been meditating for over 50 years, and while this meditation practice is key to my being happy and content in my personal life, hobbies, and work the techniques in this excellent article are something I have only taken up in the last few years. Sitting quietly just feeling grateful for quiet ‘do nothing’ time is wonderful, and just a few minutes at a time feels good.

You can buy really good Ramen ingredients online and make your own. If you are not a vegetarian and don’t mind eating pork, you can buy sealed packets containing about 2 or 3 fluid ounces of thick pork bone broth, reduced to a thick liquid. I like a lot of veggies, and buy good noodles.

My wife, who I have been married to most of my long life, has never liked vegetables. I make Ramen and she eats her veggies as if they weren’t even in the Ramen. Just a little life hack I thought I would share, it might work with kids also.

Are you using the $20/month subscription? I don't use it much, perhaps two 20-minute sessions a day so I don't get close to using my quota.

When it works, it is a great system, but now three times in a week it wasted some of my time.

This is my experience also: DeepSeek v4 flash is good enough for most of my work and I like the fast response times. I buy tokens mostly from FireWorks.ai in the US, but I also prepaid for a large chunk of tokens directoy with DeepSeek.

I use OpenCode mostly (uses fewer tokens than Claude Code) and I am looking forward to the release of DeepSeek’s own coding harness.

I toggle back and forth between deepseek v 4 flash/pro on FireWorks.ai using OpenCode. Easy to toggle, I default to flash.

I prepaid for a ton of DeepSeek v4 flash tokens. When I use them I would like to try purchasing Luna tokens for a while. I like cheap and fast models and since I am retired, if I waste time having to sometimes manually switch to a stronger model, that is OK with me.

> I've found that I get more done the less time I'm at the computer.

Right on! For 15 years I worked for two companies that had their offices on the cliffs overlooking the ocean in La Jolla CA. I would spend a lot of time walking along the cliffs or in the park thinking about work, and for serious work discussions I would urge my colleagues to walk with me if we didn’t need a white board.

Listen to Rich Hickey’s ‘hammock talk’ again…

Very good, pairs well with the linked article.

I am an old man (mid-70s) and retired now except for personal research and writing, but fifty years in the business taught me to simply pay attention to what people actually get done.

This looks really good, but for me the timing of seeing agent-manager is bad: I read about the similar Herdr on HN several days ago and I have completely changed my dev setup for the better.

I have always been a traditional simple command line tool kind of developer (I am an old man, old habbits die hard!) but I recommend at least a one day experiment with tools like agent-manager and/or Herdr that integrate tools like Claude Code and OpenCode into an integrated work environment.

I have been hacking on my own Emacs based agentic coding and work environment (so much fun!) but my recent experience with Herdr has been much more productive.

Anyway, the docs for agent-manager look good, I look forward to reading HNer’s experience with it.

+1 upvote

I don’t agree with you but your comment is very well thought out and doesn’t deserve downvotes.

I used to do this to build command line utilities in SBCL, with millisecond initialization times. Good technique, and compress the image so executables are fairly small.

Clog is very cool. As an alternative, I have recently been building Common Lisp UIs with webview (I have two example webview apps as recent additions to my Common Lisp book).

re: heap based delivery: not a good idea. But, I sometimes do heap based development when I have a ton of data I want loaded every dev session, then I save a SBCL image, and restart my Lisp environment using my custom image.

To me, the most weasel words are “”All sufficiently capable models, open and closed, should go through mandatory safety testing.””

I bet via regulatory capture that evaluating Chinese models will be a very slow process if Dario gets his way.

I enjoy using Opus, but I am in the process of ripping it out of my toolkit and leaving it on the ground behind me, as I walk away.

Welcome to our political system.

There are ways to fix this, in theory, but our government has passed the point of no return with corruption.

There is one solution: no one register as a democrat or republican: this would make the dirty D.C. system a little harder to corrupt the system.

Unfortunately, people let their emotions guide them in thinking about political parties. Most people are idiots who can’t see how they are divided into two groups and manipulated.

Too bad, this used to be a nice little country we had.

I agree, then need to bring back Nelson Minar’s old search API, or something like it.

Google does supply search grounding with Gemini API calls, and that is handy, but not general enough.

I more or less agree. When Apple starts using Chinese and expanded South Korean production memory chips, their hardware pricing problem will go away.

I am running new betas for macOS/iOS/iPadOS and Siri is actually useful for a much wider set of use cases. I asked Siri last week what models it was using and one of those listed was Gemini which is confusing because I enabled free use of OpenAI in the settings. Regardless, Siri is much more useful than it used to be.

Great show of principles and character, +1000 to the author.

What can we "regular people" do?

My approach, which is very inconvenient, is to try as much as possible to not feed my consumer spending money to tech giants. In some cases I fail at this (e.g., my wife and I will always buy a family membership for YouTube Plus with music; we still order some things online from Amazon) but I make an effort.

I have been an AI practitioner for over 40 years and I want to use the very best AI coding agents, etc., but I am forcing myself to just use access to cheaper open weight models from smaller US inferencing companies (my last $$ subscription runs out in a few days).

That sounds reasonable, Intel Macs aren't much used anymore.

My problem with AI use of em-dashes is that it covers up sloppy writing. I simply collected about six examples of horrible use of em-dashes and how I would have written the text. This became part of my SKILL for general writing (along with a lot of other specific examples).

I am repeating myself: at least a partial solution is enforcing anti-dumping laws, that is, make money losing subscriptions illegal. If businesses and individuals pay actual token costs, then they will naturally only use tokens in ways that make business sense.

Usually I applaud optimism, but the current approach in the USA will probably adversely affect most non-rich people, and I would like to see more public well grounded in reality pessimism.

This seems like a smart move, given their ability to host efficiently. I approve of efforts to make the cost of inference for smaller useful models slowly approach 'close to zero' and there are many good paths for getting there. It is useful for companies to get fast hosting for the class of smaller models they may end up hosting in house.

This is a real problem, a long time in the making. Over fifty years ago advertising started to affect the way people thought: I argue that they trained people to not be content with a classic simple life. Fast forward to the best tech focused on messing with people's attention. I like Center for Humane Technology for discussions of tech that is designed to help people, not control them.

Why? The current administration is doing such a fine job protecting the investments of a few hundred wealthy people, and not caring about the economy of the country.

The combination of a wide range of model sizes and capabilities, with good engineering choices as which to use for what applications, that is the way.

+1 I have read most of Rand's works. She grew up in the Soviet Union and hated collectivism and carried it very much too far. Still I find value in her work, and if she were here today I would bet $100 that she would disapprove of the current administration.

Great 'scratch my own itch' story and I like the framing of Emacs and Elisp as Malleable Software.

I share the cynicism here about this article. I think the saddest part of betting our economy on AI is that Google is actually the US-based corporation that is best positioned because so much of their AI efforts are not specifically aimed at software development. At least for now their profit margins are high enough to withstand the high AI costs, but let's wait and see if that lasts.

For me, poolside.ai “came out of nowhere” a week or so ago when I discovered their local coding harness ‘pool’ and their smaller 33G MOE model that runs fast and is effective on my old 32G mac mini. Really good work!!

I need to evaluate their large hosted model.