time0ut

I am working on a domain specific agent that includes the concept of skills. I only allow one to be active at a time to reduce the chances for conflicting instructions. I use a small sub-agent to select/maintain/change the active skill at the start of each turn. It uses a small fast model to match the recent conversation to a skill (or none). I tried other approaches, but for my use case this was worked well.

My model for skills is similar to this, but I extended it to have explicit use when and don’t use when examples and counter examples. This helped the small model which tended to not get the nuances of a free form text description.

A reason, at least for a period of time, was accuracy of the delivery systems. You can’t attack a hardened target with a 100kt weapon and a delivery system with a 1km CEP, for example.

Not the only reason of course.

Wow. This is alarming.

We have done a similar operation routinely on databases under pretty write intensive workloads (like 10s of thousands of inserts per second). It is so routine we have automation to adjust to planned changes in volume and do so a dozen times a month or so. It has been very robust for us. Our apps are designed for it and use AWS’s JDBC wrapper.

Just one more thing to worry about I guess…

My company has been intentionally causing attrition in the US by moving to effectively a 996 style schedule. As people quit, their positions are moved to the India office. It is not an officially communicated policy. I have just surmised this based on private conversations with the executives and what is actually happening.

Interesting day. I've been on an incident bridge since 3AM. Our systems have mostly recovered now with a few back office stragglers fighting for compute.

The biggest miss on our side is that, although we designed a multi-region capable application, we could not run the failover process because our security org migrated us to Identity Center and only put it in us-east-1, hard locking the entire company out of the AWS control plane. By the time we'd gotten the root credentials out of the vault, things were coming back up.

Good reminder that you are only as strong as your weakest link.

Ah thank you. I had read the article, or so I thought, but stopped too soon.

Could someone please point me to a concise summary of controversy around DHH? I have seen a few references recently, but I am out of the loop.

I have tried switching family members over after malware incidents. The most success was setting my 80 year old grandmother up with Lubuntu. She had no issue picking it up. I don’t think she even really noticed vs Windows. Lasted a few years until she went to an iPad for accessibility reasons.

I have been a Linux desktop user for 20+ years. It is incredible how far it has come. There is nothing Microsoft can do that will drive the normies away though. Microsoft knows this and that is why we are where we are.

Very interesting. I found myself nodding YES the whole way through the post. Something like this could lead to a large shift in how we manage infrastructure. We split terraform configs for more reasons than just splitting state of course, but something like this could make other approaches to organizing things more viable. Really cool and will be keeping an eye on this.

The market seems bad right now. Companies are offshoring everything they can and squeezing both sides.

At my company, we only hire in India now and the executives are intentionally causing "attrition" in the US by running people into the ground with demands that amount to 996 style work.

That is unfortunate. Not because I think they should have to, but because they eventually will have to if it gets big enough. Never underestimate the ability of your users to hold it wrong.

The default install only binds to loopback, so I am sure it is pretty common to just slap OLLAMA_HOST=0.0.0.0 and move on to other things. I know I did at first, but my host isn't publicly routable and I went back the same night and added IPAddressDeny/Allow rules (among other standard/easy hardening).

I love this website. It is a throwback to the old internet I grew up with. It has it all. Packed with esoteric information gathered and curated by a passionate group. Designed for desktop only with its own unique aesthetic. Not covered with ads and cookie banners and newsletter popups. I remember spending many evenings exploring such things at 33.6kbps.

On topic: I have watched every episode on TNG more than once and never noticed this. How embarrassing!

For me, it just requires different genres of music.

Brainstorming with LLM? Literal background noise like distant thunderstorm.

Pairing with LLM? Classical.

Solo? Still technical death metal.

The average person has no idea how industrialized fraud has become. There are essentially entire companies with management, support staff, IT, and office buildings whose business is fraud. It is sad when you think about how many resources have been poured into it.

I always seem to end up dual booting Windows to play games. I’ve made it quite a while without installing Windows now. Mostly because I haven’t had the desire to game much, but also because of how good gaming on Linux is now. I will hold out until there is some big multiplayer game with anticheat that my friends get me into.

Cool. I want a do it all python tool. Ruff is great. Having it out of the box with uv is great. Less crap to mess with. I haven’t tried ty yet, but looking forward to not having to mess with pyright.

Wary of the vc funded aspect though…

I have had the same experience within the last 18 months. The storage team came back to me and asked me to spread my ultra high throughput write workload across 52 (A-Za-z) prefixes and then they pre-partitioned the bucket for me.

S3 will automatically do this over time now, but I think there are/were edge cases still. I definitely hit one and experienced throttling at peak load until we made the change.

You can provision gateway endpoints for S3 and DynamoDB. They are free and considered best practice. They are opt-in though, but easy to enable.

Honestly, everything I have written in markdown files as AI context fodder is stuff that I write down for human contributors anyway. Or at least stuff I want to always write down, but maybe only halfway do. The difference now is it is actually being read, seemingly understood, and often followed!

Medical marijuana dispensaries are not covered entities under HIPAA [0]. The way the law works is weird, but they are not required to comply. All the more reason the federal government needs to catch up with the times on cannabis.

[0] https://www.hhs.gov/hipaa/for-professionals/covered-entities...

I have so much nostalgia for that time period.

Counter-strike was my introduction to how the Internet and TCP/IP worked. I built my first PC to play it. I learned linux to run servers for it. It inspired me and my friends to learn C to try and make our own mod. I made a website for my clan, self hosted it, and registered a domain for it.

The community was incredible, partly because of the server browser, as you point out. There was also a massive IRC community around it that was way more cohesive than what exists today. So CS was also my on ramp to IRC and the technology communities there.

I don't play a lot of games any more. Every now and then I'll try something. I have the GPU anyway and everything works great on Linux now. I found out there are third party server browsers for CS2 with modded servers. It is so tiny compared to the old days, but they exist. I played around on a couple around a year ago and had a good time. If you are feeling nostalgic, you should check it out.

Oh wow. I thought it was anthropic even after skimming the site. The color scheme and styling is pretty similar. That feels a little slimy.

If it is a higher up, someone I am actively working with, or someone I know well, then I take the "quick call". Otherwise, I push back and ask them to write out their question somehow.

This ends up a few outcomes, usually positive:

  - They give up (pretty common)
  - Writing it down helps them to answer the question themselves
  - I can directly answer with a response or link to the relevant docs
  - We have an actual agenda for the meeting they want to have

It hurt to read this. I have seen all of this and more.

  - Teams that produce negative output for years with no consequence
  - Six figure monthly AWS bills on unused resources
  - Technical people who can't use a computer
  - Constant re-orgs and turn over

Wait until this guy experiences the wrath of big consultants...

It is hell, but it pays. I get my fulfillment building my own things outside of work and dream of the day I can escape.

You made a mistake there. 2 + 2 is 5.

<insert ridiculous sequence of nonsense CoT>

You are absolutely right!…

I love the tool, but keeping on track is an art.

Using claude code feels like pairing with another programmer. Cursor feels like a polished extension of the IDE. They are both good tools and easily worth $20/mo. I think Anthropic has a 7 day free trial going on. Worth trying it out.

Thank you for sharing these prompts. These are excellent.

I've been working with Claude Code daily for a month or so. It is quite excellent and better than the other agents I have used (Cursor, Q). This article has some good tips that echo some of the things I have learned.

Some additional thoughts:

- I like to start with an ideation session with Claude in the web console. I explain the goals of the project, work through high level domain modeling, and break the project down into milestones with a target releasable goal in mind. For a small project, this might be a couple hours of back and forth. The output of this is the first version of CLAUDE.md.

- Then I start the project with Claude Code, have it read my global CLAUDE.md and the project CLAUDE.md and start going. Each session begins this way.

- I have Claude Code update the project CLAUDE.md as it goes. I have it mark its progress through the plan as it goes. Usually, at the end of the session, I will have it rewrite a special section that contains its summary of the project, how it works, and how to navigate the code. I treat this like Claude's long term memory basically. I have found it helps a lot.

- Even with good guidelines, Claude seems to have a tendency to get ahead of itself. I like to keep it focused and build little increments as I would myself if it is something I care about. If its just some one off or prototype, I let it go crazy and churn whatever works.

The article resonates. AI coding assistants are cool and fun to use, but they just help solve a solved problem faster. The really exciting thing is exploring the new problems we can solve with this tool. It has really reignited my passion for building.