Jo McKee runs a full content and marketing operation on $20 a month in AI costs. Twelve months ago the same operation was costing her $1,000 a month, same output quality, roughly the same volume. In this solo episode of The Jo Show, she walks through the exact setup that made that change possible — not as a theoretical framework, but as the live system she runs her business on.
The AI tool market has a structure that works against solopreneurs. Each platform charges subscription rates on top of the underlying API costs, and no single platform covers the complete job. The result is a stack: one app for content, one for research, one for automation, and a manual bridging step between each of them. Switching platforms means rebuilding your processes inside someone else’s system. Most business owners end up trapped not by bad tools, but by the compounding cost of reasonable ones that don’t talk to each other.
This is a solo episode — no guest, just Jo opening up her own operational stack. She covers four things: where the cost went and how it was eliminated, how she built a knowledge base that improves without manual maintenance, a Windows feature that costs nothing and that most business owners have never opened, and why the whole setup is deliberately not tied to any single AI model or platform.
Where the $1,000/month was going
The cost wasn’t one expensive tool. It was a collection of reasonably-priced tools solving overlapping problems. Most AI platforms layer their subscription fees on top of the underlying API costs: you pay the vendor’s margin on top of what you’d pay for direct API access to the same model. Across four or five tools, that margin compounds fast. You’re not buying better AI. You’re buying someone’s dashboard.
The shift Jo made was identifying which parts of her workflow required a paid interface and which could run on direct API access. Most of the dashboard features she was paying for, including templating, prompt management, and workflow routing, can be handled outside a paid platform using configuration files she owns and controls. Direct API access to the same models — Claude, GPT-4o, or whichever performs best for a given task — costs a fraction of the subscription equivalent. The models themselves are not cheaper through a dashboard. You’re paying for the layer on top of them.
This isn’t a quality downgrade. The output from Jo’s current setup is higher than what she was getting through the dashboard tools, partly because she has more precise control over how the models are prompted and what context they receive. The $980 she cut from her monthly bill was going to interface layers, not to better results. Once she identified that, the decision to move was straightforward.
A knowledge base that improves as it works
The standard pattern with most AI tools is stateless: each session starts from zero, the model knows nothing about your business, and whatever useful output it produces disappears into a document somewhere. This creates a consistency problem across pieces of content and a recurring time cost as you re-establish context every single session. Jo’s setup addresses this with a structured knowledge base that sits between her business and her AI agents, and that knowledge base updates as the agents work.
When a prompt produces a better result than expected, or a process needs adjusting based on real output, that learning goes back into the system. The agents draw from the knowledge base at the start of each task, which means their outputs get more accurate over time without someone manually maintaining a prompt library or re-briefing the model on every run. This is functionally different from the “memory” features most commercial tools offer, which tend to store surface-level preferences rather than the operational knowledge that actually shapes output quality.
The knowledge base is entirely private. It lives in Jo’s own environment, not on a vendor’s server and not visible to anyone else’s training pipeline. For a business handling client content and proprietary strategies, that’s a concrete distinction rather than a theoretical one. The prompts and agent instructions are files she owns, formatted independently of any vendor’s system, and they move with her if she changes anything about her setup.
The Windows feature most users skip past
There is a built-in Windows feature that Jo flags as making a material difference to her setup — one that costs nothing and that most business owners have never opened. She doesn’t name it in the episode description (watch the episode for that), but the practical effect is local processing capability that most paid AI tools charge separately for. It’s something that already exists on the machine, working without any subscription.
This is the kind of thing that gets missed precisely because it isn’t being marketed. Nobody runs ads for features you already have. Commercial AI platforms are competing for your subscription dollars, not pointing you toward the native capabilities on your own machine that would reduce your dependence on their products. The outcome is that many solopreneurs pay recurring fees for platform features that replicate functionality already sitting on their hard drive, unused.
Jo walks through exactly what this feature does and how it fits into her setup in the video. If you’re running Windows and have been adding AI tools on top of it, this is worth watching before you renew anything. The gap between what most people think they need to pay for and what they already have access to turns out to be wider than expected.
Not being tied to one AI model as they keep improving
The leading AI models have changed meaningfully in the past eighteen months and will keep changing. GPT-4o was the strongest option for many tasks in early 2024. Claude has since pulled ahead for long-form writing and multi-step reasoning. Gemini performs better in specific domains. The model that produces the best output for your business today may not be the right choice in six months — and if your setup is built inside one platform’s system, changing models means migrating your entire workflow, not just updating a setting.
Jo’s setup uses models through direct API calls, which means model selection is a configuration change rather than a platform migration. She can test a new model on a specific task, compare outputs against her existing baseline, and shift without rebuilding anything. The prompts, the knowledge base, the agent instructions: they’re in files she owns, formatted in a way that isn’t tied to any vendor’s proprietary structure. The capability to follow the best available model wherever it goes is built into the setup from the start.
For solopreneurs building a stack now, this is worth factoring in before things get established. The cost of platform lock-in isn’t visible upfront — it shows up eighteen months later when you want to change something and find that your setup is sitting inside someone else’s container. Building outside of that from the start isn’t technically harder. It’s just less familiar, because fewer people are selling it.
If you want help translating this approach to your specific situation, Jo has a tailored AI implementation strategy available — a direct application of this thinking to your business, rather than a generic framework you need to interpret and adapt yourself.
Want to be a guest on The Jo Show?
If you're a business owner with a story worth sharing, Jo would love to hear from you.
Apply to be a guest here.
