AI has limitations. Don't eat the plastic. 

All the Blogs here are human scribed. No AI involved. Not even AI-edited – so any errors or dodgy grammar – that’s on me. Zero AI. I use spell-check in Word – but won’t allow Copilot to interfere. You are getting Natural Intelligence here (Far vanity reasons I will not disclose my bench-marked natural intelligence numbers – the venerable IQ). Any long-windedness is my fault. Same for chasing squirrels and getting side tracked.

Pictures in the articles are all AI-generated.

I used AI to build this site and the engine that finds and renders the articles. It has been an education in AI. I have a development background – that helped. Some days I loved coding with AI – other days I wanted to place my laptop on a busy freeway and have trucks ride over it.

I used Claude almost exclusively. I dabbled with ChatGPT, Grok, Qwen, Kimi and Copilot – but kept coming back to Claude. Claude Code launching end of 2025 was wonderful. It is awesome.

Awesome. Not perfect. It sometimes makes stuff up – the hallucination thing. I was troubleshooting a bug 4 weeks ago and it made up test results. So not perfect. AI reasoning also gets stuck in loops. The human intuitive jump is missing. Quite often it troubleshoots something and gets ‘fixated’ on a cause or solution that is not legit.

I would have needed to learn 4 different programming languages to create the site and its underlying engine. I would have needed to develop expertise in a number of other related linguistic and logic frameworks to pull the engine together. This would not have been possible as a hobby project on the side without AI.

The version that I have taken live is version 4. Fully redesigned. The earlier versions failed for various reasons. Version 1 was vibe coded. Vibe coding is a wonderful experience. Its fun, you get to create something in minutes. It’s great for prototyping or spitballing.

It used to be that you would draw a design/plan/idea on the back of a napkin. Now you can just enter a prompt into an AI and voila – just like the genie in the lamp the AI makes it real.

But its not real. Its not a fully fledged workable piece of software. Yes – I’ve read the articles about how someone drew a design for an app on a napkin, took a photo of it and pasted it into AI and then launched the app it made and made $2 million in the first 6 months weeks days. Then lived happily ever after. That’s how you end fairy tales.

Either way – vibe code – play with it. Experiment. Brainstorm. Its fun. Its creative.

For version 2 I went full waterfall monolith program. I created a High-level Requirements specification. I then broke this down into 7 discreet detailed requirements. Next was Architecture. Then I created an epic plan. Immutable contracts and handoff documents. Each section of the system had a fully-fledged SDLC.  

I had a folder of documents, a rigorous playbook. Every chat started with the overall design, then the requirement for the section. Then each sub-section of the system had a design, a test pack. Finally, there was code. It was so complex and comprehensive I started to create small process flows to keep track. It stole the joy from my life. You’ve all run that project or been on it.

Version 3 the pendulum swung back. Minimalist. Agile. On the flipchart in my office, I wrote the following mantra:

PERFECTION is the ENEMY of PROGRESS

Lean was my new watchword. Not vibe-coding, but minimalist specs. I re-read the Agile Manifesto. I pasted it into the project steering contract.

I built a lean, mean horrible machine. Horrible.

Version 4

Somewhere in the middle. I use AI for everything. I also review everything. I scrutinise and check. I challenge. I use sub-agents to validate.

AI can be long-winded. Subscriptions cost. Chats have limited context windows – long, verbose conversations eat tokens. Never mind ads in AI – I believe that the AI companies set their systems up to be purposely long-winded. LLMs are great at this. Munching your tokens like a Pacman on meth.

Newer models have bigger context windows. It helps – but its not perfect. Models lose track of context.

The point of all this? I did mention I get long-winded – then again – you’re not paying for tokens.

The site uses AI heavily. The aim is to search the deluge of articles that land on the net daily and try and find the needles in the haystacks. Needles that are at the juncture of the PMO-sphere (everything PMO related) and AI technology.

These are the content I wanted to find daily. Its very limited. There are approximately 8.5 million pieces of content landing on the net daily. Of that there is a subset devoted to AI. There is also a smaller subset devoted to PMO-sphere. There is a miniscule intersection of those 2 subsets. That is what smartPMO.ai is searching for.

Think of it this way:

SmartPMO.ai searches the daily haystack (haystacks are estimated to contain 3 to 10 million strands of grass – so this analogy is pretty good), finds a few thousand candidates and then uses AI to determine the best fit. From there we score the top we can find (there are quality thresholds) and this gets analysed by AI and lands on the site.

On a really good day there will be 20 new and fresh pieces of content. Most days you will get less than that.

Every now and then you will get a piece of wire or a piece of plastic. Sometimes the stuff in the meadow gets caught up in the haybale. In the same way sometimes the AI curators will select an article that is not relevant.

This is where humans get into the loop. Hit a vote button. This will feed into the system and will be used to keep fine-tuning the selection engine.

Once a week there is a newsletter that consolidates the 20 best pieces of content found in the past week and mails them to you. You can subscribe or just bookmark the site and return daily.

I toyed with the idea of sending a monthly – Best of the month – but at the pace at which the field of AI is evolving – reading a 30-day old article is probably outdated.