25 years designing OSS/BSS software for large North American telcos — the kind of systems that never quite achieved full automation and always needed humans in the loop just to keep them afloat. I spent two decades watching painful human involvement — business users spending noticeable effort learning complexities of UI and frameworks — persist through every wave of tooling. I want to minimize human involvement — removing people from the loop wherever automation and AI can take over — and give AI the power — under control of a power human user, the Creator — to design and implement business processes and perform quality control, bringing development to a new level of efficiency and economy. The next generation of software systems will be oriented towards machine-to-machine interaction and JSON/YAML exchange, with Human UI playing a supporting role for visibility and debugging, and will be partially predicate-based to simplify interaction. This will radically reduce the cost and time of application development, eliminate the long learning curve, and remove an entire class of human error from the equation.
I have a deep scientific background in math, physics and even prehistoric AI applications — from the era when 128KB of RAM was a luxury and semantic networks were as fragile and innocent as snowdrops in early spring.
Now we have finally reached the computing power and artificial intelligence capable of realising these aspirations — and yet we wait with a mix of excitement and apprehension for the moment when this new model of human-AI interaction lifts off the ground and takes flight. HuMaDev is my new tune — a Prolog and semantic-based framework where a human Creator and an AI Agent jointly design and implement software with formal validation at every step. We have already rolled up our sleeves — whistling Learning to Fly by Pink Floyd. Starting with an article on the Semantic Network Layer as Persistent Context between Human and AI to run AI Agile Development under Creator supervision. Next articles and GitHub coming soon. Writing about the bigger picture on Medium and in articles on this site.
HuMaDev (Human-Machine Development) is an original framework for collaborative software development between a human Creator and an AI Agent. Its core idea: a semantic intermediate layer — readable by both humans and machines — that validates whether what was built actually corresponds to what was intended.
Built on Prolog predicates, semantic frames, and a filesystem-as-graph approach. Requirements have formal representations. Test cases validate frames, not just files. Sprints are truth queries over a semantic network.
Co-developed with Claude AI. All the good ideas are mine. All the tedious parts are Claude AI's. We agreed smoothly — which is rare in the human world.
Some ideas about automating everything with AI may look radical — but even the strongest ideas in this world suffer from the same fate: they don't fit into the narrow, over-controlled pipelines run by entrenched gatekeepers. This is the main reason for starting this site.
I spent a couple of years before 2000 as Editor of the Networking section at PCWeek/RE — so I know what it means to find a signal in the noise. If you want to share smart ideas or thoughtful commentary for the HuMaDev project, email me. I'll review the best ideas and opinions and consider them for articles here.
Some documents and code will appear on GitHub. Joint projects can be considered if they bring meaningful support to the project.