Paying To Play

Usable Artificial Intelligence will NOT remain low cost.

The startup I am involved in is really putting me through my paces. We have daily stand ups, in person meetings, and I have a pack of new technologies to master.

Things in my queue include supporting people using Figma, getting swift enough to interview React Native developers, I have to climb the learning curves for both Letta and MindsDB, and getting some local LLMs going and making them available using Tailscale.

Business things are coming along, but much slower than I’d like. Last month I had a choice of the cell phone plan I’ve had for a while, OR my ChatGPT ($20/month) and Claude Code ($20/month) subscriptions.

That took no thought at all. I simply can not do what must be done without that level of access. Knowing that the AI bubble is going to burst, I expect that $20 won’t buy much a year from now, and I’m scrambling to get ahead of this.

Attention Conservation Notice:

Where is AI going in terms of careers and costs? If you have not found a way to get some value out of the entry level paid services, you are probably being left in the dust. Come on in if you’d like to know what I’m seeing.

AI News & Strategy Daily:

I can’t stop watching Nate B. Jones. I literally did some upgrades to Parabeagle specifically because I can’t take it all in at once, so I’m going to make a document set of out his stuff and the annual State of Artificial Intelligence report.

I checked myself against the Ten Level Framework. I’m out here writing about my personal AI journey. I’ve published my first MCP server. I have multiple commercial efforts where AI is a factor. That I am writing, authoring software, and trying to turn this into a career clearly puts me into the 7 - 9 range.

The difference between a good programmer and a great one is an order of magnitude. It’s not factory piece work, this has been true since I first touched a keyboard, clear back in the late 1970s. Now it’s true with AI, but … BUT … the people who are using it, those who really get it, are extending their cognition into their systems. Nate doesn’t bring as much woo woo to the game, but that is essentially what he’s saying here. Self-upgrading humans really are a thing, and it’s only going to get more fierce as things progress.

AI is a career tsunami. Like that scene in Krakatoa, you will only survive if you face the wave head on with the throttles jammed as far forward as they’ll go.

Costs:

Eighteen months ago I had a twelve year old HP workstation with an eight year old Nvidia GPU in it, with a crap laptop from Ebay that didn’t work, and which I could not return. The rack mount systems I had were just as old as the workstation and starting to fail. The machinery was end of life and after my first encounter with COVID19 I felt like I was, too.

Then, thanks to writing here, I got pelted with Apple hardware. The M1 Pro MacBook I’m typing this on. The M1 MacAir across the room that’s my excuse for a local LLM server. The iPad where I just watched the first episode of Wednesday. And two iPhones. The generosity of those of you who read me provided a new lease on life.

The effects of this can NOT be overstated. I would never have bought a MacBook, I don’t have that kind of money, the five I’ve owned in the last fifteen years were all castoffs from clients, or these recent gifts. I would probably still be crutching that old HP along, or maybe I’d have a slightly newer Z440 so I could run LM Studio. These machines are 30+ pound boat anchors, not something I can stick in a backpack and take across town for a meeting.

So I was blessed with what proves to be marginal hardware for 2025. I had enough time and spare cash that I began exploring the entry level services, and they’re now indispensable. The 8GB Mac doing LLM duty is not sufficient for the task, so I’m in full tilt scrounge mode, trying to come up with something that offers at least 16GB just for models alone.

If I am to do the readily apparent things I must do to be successful, I can’t accomplish that with an unlimited services budget, nor could I do so with an equally generous hardware budget. If I am to steer an enterprise through the maelstrom, I am probably going to end up with an entry level Kia payment worth of services, AND a three year old Kia Soul worth of hardware either here or at the office.

You Do You:

If you are reading here at all, you have a job that involves handling information at least some of the time, and you’re going to probably want to get better at it. This means you should be at a bare minimum using one of the LLM services and getting some value of out it.

The first hurdle is that $20/month cost. I’d consider this to be an URGENT need.

I would have said at the start of 2025 that you can do this without hardware, but I don’t think that will remain the case in 2026. The MCP extensions to Claude Desktop, VSCode, or other graphical systems that will serve as clients are still a technician’s job to install. That obstacle will not be there much longer - DXT was still more theory than practice when I first tried it, but the use of such extensions will be within the reach of everyone in 2026. And if you’re going to do something like Parabeagle, your system MUST have GPU hardware acceleration for the local duties.

If you’re on Apple, do you have at least 16GB of ram? If you’re on Windows, does your GPU offer at least 8GB? If not, you know what needs to be on your Christmas list.

Project:

You need to have a project, something with some sort of objective to it, so you can focus enough to reach the goal. If you have no idea what this should be, then finding out is your project. This video from Nate is about prompts he made that are intended to help you sort out what makes sense to do.

I’ve watched this video, but only ONCE. That’s sampling for me, if I’m going to learn it’s gotta be at least twice, perhaps even three times. I will make time to do this, probably over the weekend, but that’s for the sake of communicating to you, constant reader.

I have never not been reinventing myself, this is the fate of every square peg autism spectrum adult. I have been sampling AI for some time, and when I saw that video about MCP servers early last summer, the lightbulb went off for me. Prior to that the things I tried, like Dify, just had enormous learning curves and poor onboarding.

If you come across something in your sampling process that seems to not work, maybe ask ChatGPT and/or have a look around Reddit to see if you’re not alone in frustration. Part of the sampling ritual is discarding things that either don’t work, or more likely just don’t fit for you.

Conclusion:

I hope this is the nudge you need to get moving on your own AI upskilling. Even with there being an enormous bubble that’s going to burst, there are some things of which I am confident.

• AI will not disappear, it’ll just become more costly.

• AI will be a dividing line for jobs, like computer skills were fifty years ago.

• If you are not a self starter, you should pick a couple people like Nate to follow, so you can stay abreast of changes.

• You’re going to have some recurring service costs and probably capital costs as well, unless you can land a castoff gaming computer from one of your offspring when they upgrade, or arrange for something with more punch when you get a work upgrade.

The startup I am involved in is fundraising - we’re all hustling for other work to keep us in the game until we can close. That established commercial thing that was supposed to start paying the bills around here in August is not dead, but it’s taking dramatically longer to get something done with it than I had hoped.

I kept my nose above water for third quarter by persistence and dumb luck. I do NOT have a plan for November yet and that’s more than a little scary when it’s just two weeks away. If you’ve been getting some value out of reading here, I would greatly appreciate it if you would subscribe.