Peter Wemm

Remember the old 2600Hz thing and how much money it cost AT&T? In hindsight, it was so obviously dumb to put control signalling and user-data in the exact same channel. We'd learn from that, right? It's so obviously a terrible idea that can never work safely that we'd never do something that dumb again, right? RIGHT?!

Oh wait. That's pretty much standard operating procedure with AI agents. Just jam it all in the same context, what could possibly go wrong?! Surely it'll be OK this time, right?

*Bangs head on desk*

I got a bit of joy after watching people use AI coding agents and trying to shoehorn it into their "Agile" processes. I'm starting to get my hopes up that this is a perfect storm that might actually slay Fake-Agile. Fake-Agile, you ask? Every company that I've seen that implements "Agile" has a wildly different thing that has little in common with what Agile set out to achieve. It's a different beast entirely that seems more optimized to delay and diffuse responsibility than speed up. Fake-Agile is an abomination.

And now, people are absolutely swamping it with AI-generated stuff. I've seen AI-generated tickets, leading to AI-generated user-stories, leading to AI generated code, leading to AI-generated reviews, and of course AI-generated approvals. It seems like there's a 10x increase in volume and I don't think many people actually read it any more. People are even starting to not even bother to pretend.

Maybe the facade can be just torn down? Just(tm) pair-program with an AI stack and be done with BS.

Nah, it'll never happen. Managers like their metrics too much, and developers develop metrics rather than products.

Since #ZFS snapshots came up as a #sysadmin helper I figured I'd share something that has saved my butt many, many times in #FreeBSD over the years.

If you're going to in-place update a long running service, even when you do a before-and-after zfs snapshot, ALWAYS make sure the service restarts fine before you touch anything. ALWAYS.

There are not many things worse than doing a snapshot, update, spending hours trying to get a service to start and rolling back bit by bit before finally giving up and restoring the snapshot - only to discover that the service had silently broken weeks ago and wouldn't have started cleanly at all no matter what. Only update a verified-working system.

Its right up there next to some of my other personal golden rules:
* if you're going to remotely administer a machine, do the install remotely even if its temporarily on your desk. Find out on day 1 if a reinstall will be a problem.
* When you touch a 24x7 system, always make sure it comes up cleanly after a reboot or power cycle with hands off. If you had to fix something during the boot, then test it again. Repeat until it comes up cleanly, hands-off.

I know - containers, cloud, CI/CD, etc, etc. Let me have my moment!

It’s been a long day - electrical work in the attic and scary high ladders to the second floor ceiling. But the supervisor and I are OK.

According to local lore, she was left behind in 2010 when her family moved out from a nearby house. She was skittish and timid, but I helped one of the other neighbors look out for her and make sure she got enough food/shelter/warmth. Our other cats were mean to her so bringing her inside wasn't really an option. She was coping reasonably well, all things considered. I saw her chase a fox away a few months ago. However, this winter was hard for her. She must have been about 20 years old. I brought her inside anyway when a cold snap/rain/freeze was coming. At first she seemed to pick up but it became obvious that her time was up. Her kidneys finally gave out. At least she was comfortable before her final trip to the vet. She wasn't a direct part of the family but we got attached anyway after having spent a few years with her.

The ceiling fan is fixed. It had one of those stupid fan/light canopy remotes. It failed for no obvious reason. And seems unrelated to the other electrical work I was doing. It’s gone with extreme prejudice and replaced with a separate light/fan controller.

Ladder was scary though.

Today, while having ADS-B exchange open on another monitor - a plane popped up that I hadn't seen before - the Boeing E-6B Mercury. Today I learned about the role of these planes. Hrmmmm.

Presumably they wanted to be noticed as they switched transponders on mid-flight.

They were more prevalent in the cold war but still .. their adjacency to nuclear strikes, contacting nuclear submarines, and doomsday response is .. unsettling.

I should take another look at this. Tayga does a bunch of internal packet level impedance matching beyond simply rewriting the header. Presumably af-to takes care of all of that unpleasant conversion stuff?

I expect it would be relatively simple plugin - eg: the diff and 3-way merge code is already in there - see libpkg/diff.c, libpkg/merge3.c. A bit of plugin scaffolding, sqlite glue, how hard could it be?

As a followup: there was a problem. I just couldn't see it. I spent a few frustrating hours going over the details, tracing stuff, etc. It turns out the problem was a PEBKAC issue. Somebody(TM) mixed up the stickers on the wifi APs so that wifi-3 and wifi-4 were reversed and Somebody(TM) was debugging the wrong AP's connectivity. Which caused an embarrassingly long time to discover an AP was plugged into ether18 instead of ether17. Which didn't have the CAP OOB vlan present. Needless to say, the AIs were absolutely useless and actually were a source of wild goose chases. Note to self: do more stuff with lldp/cdp etc.

I would have thought so too, but.. it seems unrelated. Maybe I'm thinking about it wrong? Wouldn't the ultimate manifestation of PWMF be that it failed because of unnatural natural causes but in such a way to make me think it was something I did? Yeah, that would fit.

Sure, but those are for people with the right accent etc. Walking around with a "kick me!" sign on is significantly riskier for some.

I've had an interesting few days. I've been largely in a holding pattern while my $spouse heals from a total shoulder replacement (ouch!).

I decided to try some bug hunting in a game when I had a few spare hours. It (Stationeers) is written in a language that I'm not fluent in (C#) and has a number of irritating logic errors in it. I used dnspy to quickly decompile the relevant methods and on a whim fed that to a local LLM. But with a suitable bug report in the prompt and a few minutes of fans whining it correctly identified the cause and put together a fix. A quick edit with dnspy confirmed that it worked. While on a roll, I got it to convert its change to a Harmony patch. Which also worked first go.

It took longer to load the game, test it, and exit than to identify the bug mechanism and generate a quick fix.

Damn.

Sheesh, I've barely even started Doing Things(tm) today and I've already (slightly) injured myself - tipped an oxygen cylinder onto my foot. That wasn't even on my bingo card.

Next up: climb a 20 foot ladder to investigate a ceiling fan high above a staircase. What could possibly go wrong?