After exploring ways to have an MCP server deliver images, 3D models and animation to a chat client, I have concluded that is a really bad idea. There's too much inconsistency (even among a vendors own tools) and presents several problems for CLI-based apps. Best option: load / manage externally.

Source available! I tossed together a merge bowling shuffleboard game that you can run in your browser. It has a workflow for publishing to GitHub Pages. Source: github.com/mitchallen/g...

How the founders of Meta, Google and Yahoo(!) became rich is not rocket science: 1. Build something to attract eyeballs 2. Build your own ad engine 3. Profits The trick is to find deep pockets to keep the servers running until it becomes self-sustaining.

Instead of simple PRDs (Product Requirement Docs), I write entire handbooks. I treat AI agents like junior developers and give them a structured sequence of steps to follow. I find this keeps them on track: if it doesn’t, I refine the handbook.

My project to generate 3D printable mazes. Working out the creation of Weave Maze STL (3D Printer) files. I could sell the app (binary or micro SaaS). Or just generate and sell a bunch of models in a zip file. Or just print custom mazes and sell them as a side hobby.

An area that I’m exploring is how to use AI to build and run a zero-employee business. Something that sits quietly in the background, generating MRR (Monthly Recyrring Revenue) on autopilot. It’s not a new idea. There are a lot of people thinking about it, or actually doing it. Images ChatGPT

Here is an example of “evolving” new software programs or functions using mutation as opposed to crossover (illustrated yesterday). Instead of crossing over parts from two parents, you just clone a single parent and mutate one of the genes / subnodes. Image via ChatGPT.

Genetic Programming works by generating random functions, then breeding them via crossover / mutation. Each candidate is evaluated with a fitness function based on how closely it matches the desired output. Over many iterations, the system can evolve functions that solve the problem. Img: ChatGPT.

I’ve added a free online SVG QR Code generator to my SVG tools site. #svg #qrcode svgsvgsvg.com/svg-qr-code

More rule-based game improvements via AI. I had Claude Code run a few hundred simulations (CPU vs CPU), and White was winning every time. Then I had it analyze the situation and come up with ways to improve the chances of either side winning. Finally, Black managed to win a few. #gameengine

I extracted the core rule-based game engine from my Tic-Tac-Toe game. I’ve been adapting it to other games: Checkers, Chess, Go, Chinese Checkers, Chinese Chess and even Video Poker and BlackJack. #gameengine

In Star Trek: TNG, Moriarty was created by an open ended prompt from Geordi: “Create an opponent that can beat Data.” … oops …

I'm not off to a good start with "agy" (Google Antgravity CLI). It tells me that I've exhausted my quota in 10 minutes, and their progress bar is confusing.

My latest book, Browser Tool Sites, is available on Kindle! www.amazon.com/dp/B0H1NHXVVM A hands-on example of how to build & deploy a simple web-based tool. I include how to use an AGENTS(.md) file to update the tool using AI. The book discusses strategies for hosting, ranking & monetization.

When designing a nano SaaS, my new book describes 4 critical pillars. They are: - The one-sentence description of your product - The one user persona that it is for - The one outcome it delivers - The one reason someone would pay for it instead of using a workaround

One thing holding Bluesky back is its inability to import simple videos from my phone. Having to chase down a converter, upload, convert, download, move to it to my phone so I can post it - adds too much friction. One thing that will kill any app is usability friction.

Oh here we go again. Apple torpedoed one of my apps again and never bothered to tell me why. I was thinking of closing my account since I have no time for updates anyway. #apple #indiedev

To avoid posting quotes that were hallucinated by one AI agent, I've built another to review each quote and assign a status, along with an explanation. In this case, the agent found that the original attribution was misidentified. #aiagents #workflow #verification #citation #n8n #anthropic #gemini

An example of using an AI agent fallback model in n8n. I've exceeded my limit using Gemini, so the model falls back to Anthropic (Claude). #n8n #gemini #anthropic #claude #slackbot #workflow #automation

My Gemini-driven Slack engineering quote bot experiment has a pre-occupation with people named Vance. I’ll have to add quality control to the workflow or prompt. #gemini #slack #n8n #automation #workflow #slackbot

#midnight #catvideos #funny #humor #genai #cat

I generated a quick SQA explainer video clip using Veo. Even with an open ended prompt the holistic view came through. #veo #sqa #genai #explainer #sdet

Experimenting with Veo 3.1. The problem is that an open ended prompt - "explain what an AI Engineer does" - left the door open to an over the top explanation. Or this is the first sign of AI's planned takeover. #veo3 #gemini #vertexai #videogen #genai

Using Gemini Nano Banana 🍌 to extract an image from one of my photos. #wasp #paperwasp #gemini #nanobanana

I’ve made a career building code gen and low / no code tools. I get how AI based code generators and the other tools work. I thought the field was crowded. But if you niche-down, I think there’s still plenty of opportunities. #scriptable #locode #lowcode #nocode #codegen #cursor

#vibecoding

Can you spot the error that Gemini-CLI made? #gemini #geminicli #nodejs #google

Here is the prompt that I used to generate that image: “Create a chart showing a SaaS talking to a browser vs an MCP talking to a chat / AI prompt” Bluesky could use a credential popup like they have on LinkedIn.

Instead of creating a SaaS accessed through a browser, consider building an MCP accessed via a prompt. The tricky part is figuring out how to monetize it. #saas #microsaas #mcp