AI versus Marshy - Uninvited guests appearing in FB groups

I’ve recently finished a major project and I’m now looking to explore new ways to build my business. As part of this, I’ve been investigating the potential of ‘AI machines’ that can do more complex tasks. In this article, I’ll be sharing some of my findings and thoughts on this topic. Let’s dive in. The generated American flag imagery. I can understand why an AI might run with an American flag when the colours red, white, and blue are employed. But that doesn’t make it right. It felt putrid. Gross. It reminded me of an IRL example of an American showing up and just annoying me out of the blue. I was on holiday in Thailand and had been for a run before boarding one of the boats to get off the island. I was wearing a t-shirt by The Oatmeal that read “I do not believe in the wall, I believe in the blerch” ​ Preach! Cutting through my post-workout fugue was an irritating US accent: “Is that t-shirt talking about OUR wall?” At the time, Trump was crapping on about a wall. Me : “Ahh, no. It’s from a comic artist I really like and is about running.” Them : “Oh.” Well this week the online equivalent of that just happened in my football group. Building AI machines that do things I finished a major project last week and have a lot of energy and curiosity about things I can be doing. Something I’ve been exploring of late is building “AI machines” that do more complex tasks. The current capabilities are better than ever, and if you know what you’re doing (or can figure it out), there’s gains to be made. First, here’s an overview of the things needed: Workflow builder - these are tools that let you build “recipes” for connecting software and software functions together. It can be Make.com, n8n, Gumloop, or even Zapier ​ AI coding tool - this is what “vibe coding” refers too, I have Lovable, but there’s Bolt, Cursor, and others out there too ​ Scraping tool - tools that let you scrape/view the web as an agent, software or script ​ LLM API - access to any of the popular generative AI systems with an “Application Programming Interface” (API) - enabling you to feed data in to get a result, rather than prompts/messages/manual typing ​ MCPs - this stands for Model Context Protocol - these are sets of instructions that let an LLM (or LLM API) interact with a specific software without the need to set it up every time. ​ For example, a Zapier MCP lets a workflow point to it and get access to everything Zapier has access to without the need to build it/code it There’s a growing hype community of these capabilities on YouTube like Voice AI calller , Video analyser , and 24/7 Sales agent . I have sat through and watched an inordinate amount of these videos so you don’t have to. You can take these with a grain of salt. The vast majority of these are excited young people babbling about the possibilities of what they can do. What they lack is connection to real-world business challenges and problems that need solving. What I’ve been working on is connecting that knowledge and capability to those problems. A couple of friends came to me with these problems and I’ve been building proof of concepts for solving them. One is in the graduate recruitment world, and another is focused on how certain public companies commmunicate their results. ​ The beauty of this approach is that even if both of these opportunities fizzle, there’s merit to stress-testing them with the capabilities that exist right now. A common task I’ve done as a freelancer is build digital strategies - dozens and dozens of them. They follow a process: Data gathering, research, and interviews Findings and feedback Strategy document and execution plan A lot these same processes can be fed into workflows, LLM APIs, the web, and produce output that’s comparable/similar to what I’ve produced, and become even more powerful with expert-level oversight. These decks can now be pieced together with LLMs, Deep Research mode, Scraping, image generators, and human expertise So I’ll be investigating this rigorously over the next few months (and uncovering more business problems to solve) as I look to find new ways to build my business post-major project. If you’ve got thoughts, comments, or questions about this let me know. – Thanks, -Marshy p.s. I suspected this would be a long one, one thing I’m also thinking about is creating a tool for splitting up this content into smaller pieces for my website, Instagram, LinkedIn, and YouTube I recently met a founder who has built a sound machine for measuring crop readiness in watermelons, I couldn’t NOT share my tattoo with him Originally appeared in newsletter : AI versus Marshy 60 - comeback edition