Notes on systems, sensing & intelligence
Essays on building real-world AI and sensing platforms, human–AI interfaces, and what's missing between today's tools and tomorrow's systems.
From Models to Systems: The Real Bottleneck in AI
Organizations keep trying to fix system-level problems with better models, and keep being disappointed. The bottleneck in deploying AI isn't model quality — it's everything around the model: data, state, integration, reliability, feedback, and the human seams. A note for leaders on where AI deployments actually fail.
If Cognition Gets Automated, How Does Society Retrain Itself?
The first wave of AI-and-jobs discourse asks whether machines will replace us. The deeper question is what happens to the institutions built to produce workers for a world that's disappearing — schools designed in the industrial era, licenses that anchor liability to humans, identities organized around job titles. Part two of a two-part series on the economics of automation.
Will AI Collapse the Economy? Following the Circular Flow
The loudest prediction of our moment — that AI will erase most white-collar jobs — is usually delivered without a single line of economic reasoning behind it. So let's actually trace the flows. If you remove labor income from the circular flow, what happens to demand, to prices, to the whole loop? Part one of a two-part series on the economics nobody is doing.
Why the Next iPhone Won't Look Like a Startup
The advice that makes a SaaS company succeed can make a frontier-technology company fail. Great platforms look messy early, product-market-fit is the wrong yardstick before the platform exists, and the durable companies optimize for the right abstraction over near-term metrics. A case for building unfashionably.
The Text-Only AI Fallacy
Language is a compression of human experience, not a substitute for it. Systems that learn only from text inherit a ceiling: they know what we've written down, not what the world is actually doing. The path past that ceiling runs through sensing, signals, time, and embodiment — the world is the missing dataset.
Intelligence Is Becoming Infrastructure
Every general-purpose technology follows the same arc: it starts as a product you buy and ends as a utility you assume. Electricity, compute, and the operating system all made that journey. Intelligence is making it now — and that shift, from apps to platforms, changes how companies should be built.
Why Most 'Agentic AI' Systems Will Fail
Chaining language models together is not the same as building an autonomous agent. Real autonomy requires feedback, grounding, world models, and action constraints — systems, not prompts. Here's why most of today's 'agents' are built on the wrong abstraction, and what the ones that survive will have in common.
Orchestrating Agentic AI When Errors Are Expensive
Chaining LLM calls is not the same as building an autonomous system. What it actually takes to run multi-step, asynchronous agent workflows reliably in high-stakes settings — and why orchestration, not the model, is the hard part.
Voice Is the Hardest Easy Interface
Voice feels like the most natural way to talk to an AI — and that intuition is exactly what makes it so demanding to build. Notes from building real-time, high-fidelity voice agents at Althea for high-stakes healthcare settings.
Acousto-Encephalography: Reading the Brain Without Opening the Skull
A look at AEG — a non-invasive sensing approach we developed to estimate intracranial pressure and cerebral blood flow from ultrasound and multimodal signals — and what it taught me about inferring hidden states from deeply confounded measurements.
The Closed Loop Is the Product
Most of what we call 'AI' is still open-loop: it produces an output and stops. The systems that matter sense, represent, reason, and act in a loop with the world — and that loop, not the model, is where the hard engineering lives.
So You Want to Fix Healthcare? Here's the Lay of the Land for First-Time Founders
A field guide to the U.S. healthcare ecosystem for builders entering the space — the major stakeholders and what makes them tick, the brutal sales cycles, the integration tax, AI readiness, and the compliance beast. Lessons learned the hard way building Althea.
Agentic AI in Healthcare: Bridging the Gap Between Knowing and Doing
Agentic AI empowers intelligent, autonomous agents to not just analyze data but act on it — transforming patient care with proactive, real-time decision-making. A look at what agentic AI means for healthcare, where the opportunities are, and the ethical and technical hurdles that come with them.