Simone Macario

I built an HCI Scorer: a quick way to assess human intellectual contribution in research text.

The real shift is this: it’s easy to make AI systems generate output. It’s much harder to make them speak only when it matters. That’s where I am seeing leverage coming from.

The second layer is control. I am investing in clarity, including rules for when the system should act and when it should escalate. Without this, adding autonomy just increases risk instead of usefulness.

Most AI systems produce too much output and too little value. The real fix I am testing is stricter filters: higher thresholds, no noise, and strict deduplication.

This is the shift most people miss: LLMs shouldn’t be workers. They should be decision engines. If your LLM is parsing RSS feeds, you’ve already lost!

Why this compounds: Every new workflow inherits the same structure → testable → predictable → near-zero token waste No more: “agent forgot to deduplicate today” One architecture decision → permanent leverage

I’ve been adding leverage to my AI agent stack. This week: a fundamental shift → from LLM does everything → to code does the boring, LLM does the thinking This changes everything.

I'm about to convince @romeobellon to relocate.

There is no place like Silicon Valley.

When looking at a disordered sequence of events, our brain does a remarkable job of building a beautiful narrative that holds our life’s personal story

The past is a resolved uncertainty.

I've been having such an amazing time with Claude Code I wanted you to be able to have my *exact* skill setup: Introducing gstack, which you can install just by pasting a short piece of text into your Claude code

Life is a dream with uncertainty.

If you’re working with AI-assisted writing and want a transparent rubric, try it and share feedback: humancontributionindex.com

It evaluates how much original human thinking is present in a text fragment by using the open-source HCI framework

I built an HCI Scorer: a quick way to assess human intellectual contribution in research text.

But we can measure human contribution.

We can’t uninvent AI.

Try scoring a paper in your field and tell us what's missing: github.com/humancontribut…

HCI scores research across 5 dimensions: → Conceptual Direction → Creative Synthesis → Critical Judgment → Ethical Reasoning → Scholarly Voice Each scored 1-5, weighted, and adjusted for AI dependency. Use AI as a tool? Fine. Substitute AI for thinking? That shows. Open source. MIT licensed.

Just open-sourced a framework for measuring how much of a research paper is genuinely human. AI detection asks "did a machine write this?" The Human Contribution Index asks "how deeply did a human think about this?"