Practical AI, Honestly

Scott shares his experience of DOT, Muse, and Grokbot finding different roles: subject-specific bots, social connections, and business-focused help. He also describes his personal experiment with an investing bot connected to Robinhood and his effort to connect all three assistants through email. Coordination remains a work in progress. Trading involves real financial risk, and automation does not remove it.

Try it: Describe an assistant’s role in one task, then name one thing you’ll check in its output. If another tool takes the next step, note what context it needs.

Join the discussion: Which assistant do you use for which task, and how do you judge whether it helped? Keep examples general.

Previous article: https://practicalaihonestly.substack.com/p/why-i-started-taking-local-ai-seriously

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit practicalaihonestly.substack.com

Privacy and cost led Scott to explore local AI. Moving his roundtable from cloud services toward a hybrid setup on the MacBook Pro brought new questions about model roles, visibility, and answer quality.

In this episode:

* The Mini and Pro roundtable lineups and their different roles.

* Why Claude review and Codex moderation still send supplied material to cloud services.

* The local harness and AI Neural Gauge activity integration.

* A candidate that completed the tests but wasn’t promoted.

* A small exercise using harmless material and an answer you can check.

Scott hosts. RDC Software Works is the studio.

Cost savings and equal review quality remain unproven; the Gauge presentation work is unfinished. No medical, financial, employer, or confidential test details are included.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit practicalaihonestly.substack.com

ChatGPT helped Scott think through ideas and design. Claude helped with the coding. Scott carried prompts to Uno, shared terminal results back with ChatGPT, and decided what to do with the suggested corrections.

In this episode of Practical AI, Honestly, Scott looks at the work hidden in those handoffs: carrying context, checking results, and choosing the next step. Another model’s opinion can help, but the check still needs to fit the task.

Do this today: Pick one recurring AI task. Write down the first request, the information you pass along, the result you inspect, and the point where you decide whether to continue.

Host: Scott Ray. Studio: RDC Software Works.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit practicalaihonestly.substack.com

I am not a developer. I am not a programmer. I do not have a computer science degree.

That is important to this story, because I did not come to Claude Code looking for a faster way to do a job I already knew how to do.

I came to it trying to understand what the computer was doing while I learned how to make these AI tools useful in my everyday life.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit practicalaihonestly.substack.com

Scott explains the personal meaning behind Uno’s name, then revisits his January 2026 Linux experiment and February move to a Mac mini. The migration footage captures setup, errors, and retries—and raises a practical question: what should an AI assistant be allowed to do without asking first?

Show notes

* The beloved one-eyed cat behind Uno’s name.

* Starting an OpenClaw experiment in a Linux virtual machine.

* Moving saved context and tools to the Mac mini.

* Installation steps and competing Telegram connections.

* Three access decisions to make before connecting an AI tool.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit practicalaihonestly.substack.com

The earliest period Scott can reliably trace is 2024, when he was trying to gain control of the next chapter of his life and using ChatGPT to work through consequential questions. He is not a developer or programmer and does not have a computer science degree; the practical part of this journey includes learning the terminology, understanding what the tools are doing, and figuring out how to ask for the intended result. Beginning with TikTok and YouTube curiosity and a familiar habit of treating ChatGPT like a search box, he explains how that use widened into documents, explanations, computer questions, software comparisons, and better preparation for conversations with qualified professionals—without handing the tool professional judgment. The episode ends with a low-stakes exercise for turning one answer into a better, verified question.Current ChatGPT descriptions are separated from Scott's historical recollection. The 2024 account is retrospective memory rather than a verified transcript, and personal examples are intentionally generalized.Show notesCanonical article: https://practicalaihonestly.substack.com/p/when-chatgpt-stopped-feeling-likeUse ChatGPT — official OpenAI documentation: https://learn.chatgpt.com/docs/use-chatgptPrompting — official OpenAI documentation: https://learn.chatgpt.com/docs/promptingProgram home: https://practicalaihonestly.substack.com/podcast

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit practicalaihonestly.substack.com