Posts on FB/Tech

Well this sucks

Mon, 06 Jul 2026 09:00:31 -0500

I cannot believe what a roller coaster of emotions I had yesterday, July 5th.

I got a page from PagerDuty shortly after noon saying my cloud budget was reaching 50%

I started to explore AWS to start - I didn’t even think it could be Google GCP. I had about $800 in credits queued in that account and shouldn’t see spend for a few months.

But indeed it was Google. Over the next hour I was madly trying to find the culprit. I knew the project (myAnthosProject2), my oldest GCP project and the one I use in blogs the most. I could see from billing it was coming from Vertex AI but I hadn’t done any recent posts.

None of my systems would account for that kind of massive spend. I had not written any recent Vertex AI articles.

As I rotated keys and disabled things, I still saw the spend continuing. I finally (around 1:15p or so) just deleted the project altogether.

I saw at this point, the prior night late in the evening Google detected abnormal usage but had done nothing to stop it…

Meanwhile, it charged my credit card $100 and kept going.. the spend just kept climbing - against a project that was now completely deleted

I couldn’t get a person , the only billing support was an automated Chat bot - no person to find. The chat bot just kept telling me “well, you did all the things - account disabled and removed.. billing can lag 2-5 hours”.

2 to 5 hours - that is a firehose of costs for a regular person. It crawled all the way up to $210 by about 6:30p when it stopped growing.

Around 4:25p that night, in sheer desperation I started to post to my GDE channel

Okay - such torture today. the long story/version. I keep watching the monies go up and rollover as it dings my credit card another $100.. I’m hopping it should slow down soon but im so stuck. In short, an API key somehow was found out (tho searching, i never exposed it in writing, but whose to say they didnt get into my GCP account last week, view it, then go away). Starting July 4th evening someone in southeast asia (by the anomaly logs of trying to use AI keys in that region) started to hammer an unrestricted Vertex AI key.. it burned through about 850$ or so in credits i still had before it started to auto-charge my credit card. The failures in the system (imho) were i got an anomaly email at 11:47p at night saying ‘hey, you spent 515.85 in your project on Vertex AI when we expected you to spend 0.10" but it didnt stop the account, or charges or shut things off.. it let it run thru all my credits and into my card. Only when my own GCP Budget alert hit pagerduty after church and rang my phone did i come down to my office to see why i had any charges in GCP. by that point it had billed my card 100, and started fresh again. and i cannot disable credit cards (must have 1 primary), deleting the project all together after expiring the account and unlinking it still shows growing charges.. i chatted with the billing ai bot (there is just no way to reach a person) tells me i did all the things i can do and charges can queue for 2-5 hours.. so i just sit here watching it grow and grow with nothing i can do and no one i can talk to. I thought maybe i could apply for cloud credits as a GDE - i have before and i do have a GDG event to prep for, but that form auto declined saying the credit program is paused. so im SOL.

The week prior my project had been tagged briefly for “hijacking” but as the compliance team would not say where or how, I just had to review my billing and keys and they all seemed fine so I assumed it was because i was using old GCA/Cloud Code APIs

All Developers should be Wary

The end result here, is we (GDEs) all perked up.

One GDE shared an article they wrote, “Automated GCP Killswitch: How to Avoid an Unexpected Google Cloud Bill”

In that article they cover massive bills from students - $55k, 11k pounds, 50k Euros…

Another GDE linked to this TheRegister article which is pretty much my exact scenario including the issues getting details about ‘hijacked resources’. That person racked up $11k and wasn’t helped by billing (according to the article).

There is also a new Spending Caps feature in preview that might help, but its a request for private access and there are no guarantees it will be granted.

Where we sit

I didn’t sleep well last night.

I kept getting up and going back to my office in the basement to check my Cloud Budget to see that spend was done. I was still sitting on $210. This sucks, but this isn’t thousands of dollars so i need to keep perspective.

I’ve submitted some forms, and reached out to Googlers I know to see if they can put me in touch with billing, but I do not expect much.

The hardest part for me is I have proudly been a Google Cloud Expert since I transitioned from the Cloud Innovators Champions program. I’ve been less thrilled with some of the AI things lately, but I was a huge supporter and involved with Gemini CLI and Project IDX to Firebase.

However, I cannot do a program for funsies that might cost me the same as a new Buick. And if the lag in billing really is 6 hours (as I personally experienced), I just cannot trust using the AI tools unless I use a key, do a demo, then delete that key right away.

I never wanted to get paid, or make a profit talking about Cloud systems, including Google - but I don’t want it to impact my families finances. yesterday i was to the point of wanting to throw up thinking I just jeopardized our family summer vacation because a bad actor was using a hacked key.

And since I cannot know how they got in, I now have some very serious trust issues. This could mean quitting GCP and GDE all together if push comes to shove. I can’t be put into positions of doing free evangelism and then getting a massive bill.

I’ll have to share how things end up. As it stands I’m feeling like trash and loading up on coffee…

Closing the Harness and future of OS LLM CLIs

Fri, 26 Jun 2026 01:00:31 -0500

One of my favourite VS Code general use plugins for connecting to a variety of backend agent providers has been Continue.dev’s. It has releases that go back to 2024 where it has had non-stop continuous development. Over time, I have written guides here and for work on using it with Azure AI foundry and Ollama. It’s just really really good.

This week, however, seemingly out of the blue, they published 2.0.0 and marked it the final build (ever). I still was operating on a 2.1-prerelease in my systems so I was pretty sure this was a change of plans.

It would seem only recently, Cursor acquired Continue, the company, and immediately decided to halt the product and shut it all down.

Fromm the New Stack article:

The deal itself seems to have been completed around the same time SpaceX confirmed it was doling out $60 billion for Cursor, meaning that Elon Musk’s rocket juggernaut is now the owner of Continue as well. Terms of the deal haven’t been disclosed and, in truth, there hasn’t been much of an announcement to speak of. Around June 16, Continue updated its homepage with a brief message stating that it has been bought by Cursor, and an accompanying FAQ explaining that existing users have until July 15 to export their data before it’s deleted, and that recurring billing has been disabled

Thus one of my favourite plugins, from a good Open-Source company is bought by a fine LLM company, who itself was bought by a SpaceX, a company owned by the guy who ruined twitter (and most of our country by DOGE and election influence)… Dammit. Dammit Dammit.

In my other blog, I stay out of politics for the most part, but dark maga can eff right off… so as Cursor is owned by SpaceX (owned by Musky), I now won’t use Cursor and by proxy not any product of theirs.

By luck, almost as a poison pill, the great harness of continue.dev was set with an Apache 2 license - this is permissive license that enables business use. This means anyone can pick it up and run with it.

When looking about the interwebs, there was one fork suggested, PearAI submodule but there is no simple to use vsx install file and the contributing guide is a lengthy how to build writeup. There is a PearAI app but the last release is almost 14 months old so maybe it’s abandoned.

Obviously, in my toolbox I keep options, one of which has been Cline.bot

It doesn’t have the nice YAML/JSON block of config with selectors that Continue had, but it functions just fine. In fairness, the Continue.dev plugin isn’t dead (yet) but I have had to update it fairly often (every few months) for new OpenAI features presented in new LLM provider endpoints.

I contrast this to what I saw with my near and dear Gemini CLI which is now dead. Granted it’s alive for Enterprise users - but if an Open Source tool cannot be used by the common person, it’s rendered effectively dead. And the replacement, Antigravity which is functionally similar is closed source and tied, of course, to Google.

Copilot CLI is nice, but it is closed source (despite being on Github) so who knows what will happen if they sour on a vendor relationship and yank the endpoints - it’s not something you can really control.

The future

What I fear we may be seeing is a desperate attempt to close the kingdom, reinforce the walls, and create siloed solutions that can marry development to just one and only on vendor.

That is, if you like Google (Vertex AI/Gemini AI/Gemini Enterprise Agent Platform) then you go all in - and the tools just come from Google and you just deal with it. They say “no more Gemini 2.5” and you update your code or they say “this thinking model now costs more” you just pay.

Or you go all in on Microsoft and Copilot and accept if the docs are in Sharepoint, great, but if not, that might be a custom integration you may or may not be able to do. You can go all in on Anthropic, but dare you use their API with another harness, they will charge you pay-as-you-go (see Louis’s rant on it).

There are Open-Weight models dropped into Ollama but even that is a Y-combinator startup in California which could be bought and closed down as easily as Continue.

I’m not sure where we go from here. I can recall two trends that seem related.

The first was IDEs. I recall getting a less than legit copy of Borland C++ for my Mac laptop in 1997 to help do coding work. At the time, good compilers cost money and there was a trend about getting commercial IDEs. My Windows colleagues were about Visual Studio and later InterDev. My Java colleagues all used IntelliJ and Eclipse. I was doing Fortran at the time so they offered me Vi or Emacs and I did Vi. Seems as time went on, the IDEs went opensource or free. People used Notepad++ or TextEdit or a variety of other tools. For many years in the early 2010s I was quite happy with Atom. And now we see all these forks of VSCode so that vendors can load in their own AI tools - but I think the differentiator could be the tooling.

That is, if Windsurf, Cursor, Verdant, Bob, Antigravity, Copilot (by way of “VS Code”) all are about the same now, then I would see some adding good memory stores (like a Valkey cache) or better handling of OKF or Database integrations to tie to LLMs or just search setting the various tools apart. As I start to stew, maybe the times of “really great IDEs” will come back and we start to say “sure, they call can think up answers, but who has the easiest to use orchestration framework… who can help me test and review my code and make it legible.. who can tie in my data sources the best”…

The other trend that came to mind was music. We saw this once long ago with the record industry and Napster. At first it was basically free and we all got tunes. Then they tried to close the doors and some vendors (like Sony) had such proprietary hard to use formats no one used them (I have a minidisc player. i remember). Then it was commoditized with MP3s hosted everywhere (eg. allofmp3.com), then the big vendors got involved, the government played heavy hand and we were back to a few providers (iTunes), then streaming became so ubiquitous no one cared… and now my kids, who all hate AI, are into records (vinyl) and tapes.

So if I layer that trend, there is a future where our youth are hunting down O’Reilly books and collecting and trading them like the treasured tombs ([)i think of them as). I actually like this - the grand future is back to stacks of MSDN discs, big books, and local repos of text files.

It’s probably a long way off. Right now I recommend y’all download the models as they are free. Install and use tools like llama.cpp, ollama and comfy ui. Learn to host local and if worse comes to worse and everything gets egregiously overpriced or locked down, you have your mini-assistant on a nearby box.

Oh, and they still sell O’Reilly books, I recently bought some on Perl from Amazon …

you know, the online book seller.

Hello World

Fri, 19 Jun 2026 11:00:31 -0500

This is my first published post using CICD.

I am just now looking to set this up with Forgejo or Gitea.

Here is a T-Rex playing hockey: