iiinova — where your digital self keeps evolving

Conrad Ludgate @conrad.cafe · 27/09/2026

> what's beyond death Silly, we're already dead. This is what is beyond.

Conrad Ludgate @conrad.cafe · 25/09/2026

I stopped taking my elvanse (what we call it in Europe) and started taking chatgpt instead 💀 I didn't enjoy being on stimulants all the time

010

Conrad Ludgate @conrad.cafe · 20/09/2026

Told my skip manager that I actually don't want to chase a promotion right now. Felt good. Weight off my shoulders tbh

070

Conrad Ludgate @conrad.cafe · 17/09/2026

I've started to consider a single person doing something that can wake up an SRE as a kind of vulnerability. The line is ever so fuzzy though

010

Conrad Ludgate @conrad.cafe · 15/09/2026

How many times did it headbutt you while typing this

030

Conrad Ludgate @conrad.cafe · 14/09/2026

I've been reading The Every by the pool/beach every morning while going in and out of the water 🏖️🏊 The afternoon is too hot outside for me, and there's not much to do in the hotel room. I would watch something but the WiFi here sucks. I do have my steam deck but I can only play so much

000

Conrad Ludgate @conrad.cafe · 13/09/2026

Even with self hosted runners, github actions is a nightmare 😞

010

Conrad Ludgate @conrad.cafe · 13/09/2026

I'm still not home, but unfortunately modern technology allowed me to vibeslop this monstrosity from my hotel room with just my phone github.com/conradludgat...

github.com

GitHub - conradludgate/loadpace: Client side load balancing and back pressure library

Client side load balancing and back pressure library - conradludgate/loadpace

110

Conrad Ludgate @conrad.cafe · 13/09/2026

Why did I spend my vacation thinking about congestion control and client side load balancing? In other news, I got bitten by a fish in the Mediterranean Sea the other day

231

Conrad Ludgate @conrad.cafe · 13/09/2026

step 3: get your agent to write the classical automated flaky detection, and run an agent in CI that fixes them for you

110

Conrad Ludgate @conrad.cafe · 09/09/2026

Indeed everyone shouldn't write their own software - however, it seems like we can have more companies of smaller sizes maintaining lots more software. More diversity, similar quality. As well as having more choice, it's easier to now customise said software for your needs

010

Conrad Ludgate @conrad.cafe · 09/09/2026

I think this only makes sense cost-wise if you're already forced into API pricing (enterprise seats) like we are. For personal projects I'm currently giving $20/month to OpenAI and 99% of the time using the cheap Luna model which means I never hit any usage limits

010

Conrad Ludgate @conrad.cafe · 09/09/2026

If my understanding is incorrect, then something needs fixing. If a decision was made I should understand why. Currently K3 is very willing to make its own decisions and then think the decision came from a human. I make use of the Codex "approve for me" feature which occasionally flags such issues

000

Conrad Ludgate @conrad.cafe · 09/09/2026

During implementation phase, it's committing by itself and writing/running tests. While it's running I'm off doing something else. After refinement I usually use another session with another LLM and I get it to review it. I also make sure I provide my understanding of the feature to compare against

100

Conrad Ludgate @conrad.cafe · 09/09/2026

I'm mostly using K3 in Codex right now. It's pretty good. It's not cheap but it's comparable with the duopoly. I use it the same way I used Sol or Opus: 1. Sketch out the plan for the feature implementation - several prompts 2. Implement it - one prompt (+skills) 3. Refinement - many prompts

100

Conrad Ludgate @conrad.cafe · 09/09/2026

Step One: tell everyone you know that you're in a state of anticipation Step Two: accept you were not sane to begin with

010

Conrad Ludgate @conrad.cafe · 09/09/2026

I'm mostly using kimi k3 (open weights) with a bit of Sol (proprietary) on the side at work. Sol is a bit better, but I know Kimi will never stop me when I'm fixing bugs in my service, which Sol is always so happy to do (bugs in my service means security flaws)

121

Conrad Ludgate @conrad.cafe · 06/09/2026

But also I do want to help dispel all the doomerism sometimes. People think they're not gonna have any work left in 5 years time, and idk... The way I do my job has radically changed, but what I do in my job is exactly the same

030

Conrad Ludgate @conrad.cafe · 06/09/2026

Just this morning I was just advocating for more AI-hate/AI-skeptic voices in our AI channel on discord 😞 Though it's partly for practical reasons. That channel is for all AI discussions and the rest of the server forbids it. So rants should go there too, but they don't want to be with the AI bros

130

Conrad Ludgate @conrad.cafe · 06/09/2026

I'm now imagining some future where plate tectonics accelerate and you will be constantly climbing up the ever forming mountain

100

Conrad Ludgate @conrad.cafe · 06/09/2026

I'd become a baker or cheese maker, idk

220

Conrad Ludgate @conrad.cafe · 05/09/2026

Sent from my Felineeeeeeeeeeehhhhdddddddddfdddddddddddddddddjjjj

000

Conrad Ludgate @conrad.cafe · 04/09/2026

Happy anniversary my love

060

Conrad Ludgate @conrad.cafe · 04/09/2026

Write a blog post ❌ To cause drama and hate ✅ To remember why you left

110

Conrad Ludgate @conrad.cafe · 29/08/2026

Amos, I spend enough time arguing with humans on discord. I don't want to spend more time on that hellhole. Please use a more appropriate platform for arguing with LLMs, maybe Bluesky?

120

Conrad Ludgate @conrad.cafe · 29/08/2026

How can Oxide compete with this

1200

Conrad Ludgate @conrad.cafe · 28/08/2026

Sharing my chat sessions for my code feels like an invasion of privacy tbh

031

Conrad Ludgate @conrad.cafe · 27/08/2026

Hmmm. This is giving some credibility to the project I've been working on... 🤔

010

Conrad Ludgate @conrad.cafe · 25/08/2026

GPT is suggesting to me that AVIF2 (using AV2) will be about 20% file size reduction on AVIF (using AV1). Apparently JPEG is working on JPEG AI (not a joke, it uses ML to decompress) but I think efforts have stalled a bit?

120

Conrad Ludgate @conrad.cafe · 24/08/2026

Biased because they are my employer, but for enterprise I'm using Databricks AI Gateway. However I'm interested in personal AI gateways here

000

Conrad Ludgate @conrad.cafe · 24/08/2026

I'm thinking that the cost of an rpi running a gateway and the cost of hardware capable of running lots of VMs is quite different. Maybe there's a middle ground there of people who want some better security but don't want to run their own agent harnesses on their hardware. Idk.

000

Conrad Ludgate @conrad.cafe · 24/08/2026

I'm working on an interesting product idea for personal AI that involves lightweight VMs for reproducible LLM evals, but I don't want you to give your API keys to me. If your assistant goes rogue deleting emails you can turn off your gateway and everything stops. All gateway access is logged

100

Conrad Ludgate @conrad.cafe · 24/08/2026

Would someone security conscious enough to host an MCP/AI Gateway in their own home then be willing to let the agent run on a VM in the cloud? What if the machine the VM was running on had no direct internal or external Internet access, all access goes through a tunnel service via PrivateLink?

210

Conrad Ludgate @conrad.cafe · 22/08/2026

Practically I think the information can be private by default and only viewable by the account owner, treating it as a proper audit log rather than leaking more about the setup

000

Conrad Ludgate @conrad.cafe · 22/08/2026

The LLM should know that it was an intentional decision and not a mistake it made. I'm working on an experiment that will give each session their own access tokens, but fundamentally it'll require effort from the services to store some identifiers.

100

Conrad Ludgate @conrad.cafe · 22/08/2026

We require all LLM PRs opened in draft mode at first. Very often I'll mark the PR as ready for review but then I'll fix something flagged in review. The LLM will get confused that the PR is no longer a draft and it'll mark it as draft again...

100

Conrad Ludgate @conrad.cafe · 22/08/2026

I think platforms like github should invest in exposing audit information. An agent should be able to see a log of events that occurred and know whether it was me, another agent, or itself that made some change.

100

Conrad Ludgate @conrad.cafe · 21/08/2026

It will be build system agnostic, software packagers can publish nix or bazel or cargo or whatever build system tooling for the package version if they wish. They can even publish patches if they need to.

000

Conrad Ludgate @conrad.cafe · 21/08/2026

So anyway, I'm currently thinking that package sources live in an OCI registry of your choosing, and the package metadata (dependencies, version, etc) lives in your PDS. It will support off-platform dependencies via packageurl.org.

packageurl.org

Home | www.packageurl.org

Landing page for www.packageurl.org.

100

Conrad Ludgate @conrad.cafe · 20/08/2026

I will not solve supply chain attacks. You're welcome

020

Conrad Ludgate @conrad.cafe · 20/08/2026

However... Cargo-vet as an atproto lexicon just makes sense...

010

Conrad Ludgate @conrad.cafe · 20/08/2026