The Human-in-the-Loop Seam: Where Humans Belong in a Dark Factory
Human-in-the-loop is a phrase everyone uses. Almost nobody uses it correctly. The seam is in the spec, not in the implementation.
Articles on AI, leadership, cultivation, and the human side of technology.
Human-in-the-loop is a phrase everyone uses. Almost nobody uses it correctly. The seam is in the spec, not in the implementation.
A dark factory can transfer code. It cannot transfer the judgment that knew the code was right. That is the problem provenance was invented to solve.
An operator is a control loop with a domain in it. That is the whole definition. Here are the four operator patterns and the one anti-pattern to avoid.
A dark factory is a control loop. Kubernetes is the most boring control loop in production. That is the reason.
A light factory has too many gates. A dark factory has too few. The middle is where the engineering happens, and the middle is the part nobody talks about.
Every conversation with most AI systems is a first date. Here's the three-layer memory model that actually compounds — from Scott Scoble's scobleclaw experience.
Every software project has specifications, and every specification eventually lies. The spec-code gap is a provenance problem — and Speckl closes it by making the spec compile into code, so the gap never opens.
A roundup of Scott Scoble's open-source projects — Speckl (spec compiler with provenance), ScobleClaw (agentic engineering framework), and Beyond the Prompt (workshop on building with AI agents).
SQLite runs on every smartphone, every browser, every embedded system. Its core state machines have never been formally verified. Following the Kafka proof, we're tackling the most deployed database in history with SpeckDL and Z3 bounded model checking.
The phrase is everywhere. The working meaning is gone. Here is what it is, what it isn't, and the five positions on the spectrum.
Gallup 2026: 80% of the global workforce is disengaged. The $438B cost isn't a tool problem; it's a people problem. Here's the cultivation alternative.
Tall stalks, ground cover, nitrogen fixers, pollinators, deep roots. The agricultural metaphor isn't poetic; it's diagnostic. Here's how to read the soil under your team.
I started writing code in 1985. I started getting paid for it in 1996. The tech stacks changed; the human patterns didn't. Here's what three decades teaches you when you pay attention to the people, not just the code.
The anterograde amnesia metaphor for AI memory architecture. Three-layer memory model for agentic systems from Scott Scoble's scobleclaw experience.
Meta has Hatch. Google has Remy. Everyone is building agents. Whose capability compounds?
Your agent swarm runs 24/7 and you're more anxious than ever. Stanford's 2026 AI Index shows a 20% drop in junior dev employment since 2024. The agent revolution promised freedom — instead it gave us token anxiety and a new kind of burnout.
Gallup says 80% of the workforce is disengaged. The heaviest AI users are burning out fastest. The fix isn't training - it's structural.
Every company running AI adoption is making an implicit bargain: data for tools. The companies extracting the most surveillance are hemorrhaging their best talent.
Gallup says 80% of the global workforce is disengaged. The $100B+ engagement industry has failed for 20 years. The fix isn't perks or surveys; it's structural.
Companies will spend $60B+ on developer tools in 2026. But the real DevEx crisis isn't tooling — it's autonomy, mastery, and belonging. As code production becomes commoditized, the human dimensions of engineering become the only competitive advantage that matters.
The simple mentoring framework that actually gets used - three questions that unlock real growth.
The mindset shift that separates thriving teams from burned-out ones - cultivation over extraction.
A three-layer framework for understanding what makes engineering organizations healthy - or not.