AI, Attention and the Practice of Seeing Clearly
I have spent the last few years thinking in public about AI, mostly on X.
That has been valuable. X rewards speed, compression, and cleverness. It trains you to notice weak signals early, find the interesting part quickly, and say the thing before the moment has passed.
But a lot of the work I care about now needs more depth.
Model launches are not just product announcements. They are changes in what is possible for builders. They are changes in what is possible for humanity.
Benchmarks are more than just scores. They are signposts to what kinds of work models can and cannot reliably do.
Agent systems are more than shiny new software services. They are a new way of interacting with the realm of information; what Teilhard de Chardin called the Noosphere.
And underneath all of this is a deeper question I have lived with my entire life:
What is attention?
Not just transformer attention. Human attention too. The thing that determines what we notice, what we ignore, what we optimize for, what we believe, what we practice, and what and who we care about.
That is what “Attention Heads” is all about.
Attention Heads is a place for writing about artificial attention, human attention, and the practices that help us see more clearly. Deep, evidence-backed interpretation of what AI is really doing to our cognition, our agency, and what it means to be human.
Because the fulcrum of existence hangs on attention.
Most of the early writing here will stay close to AI and agentic engineering, because that is where my public work already has the most momentum and where the changes are happening fastest.
I will write about:
new AI models and why they matter
AI agents, coding tools, evals, memory, context, and harness design
research papers that matter for people building real systems
the competitive dynamics between OpenAI, Anthropic, Google, Meta, xAI, Chinese open-weights, coding-agent companies, infra providers, and open source ecosystems
practical patterns from my own work with Codex, Claude Code, Linear, GitHub, Obsidian, local tools, and long-running agent workflows and more
But I do not want the publication to be trapped inside machine-like AI commentary.
The larger territory is attention, intelligence, reality-testing, contemplative practice and living with a dynamic mind: how minds and machines perceive, model, reason, fail, recover, and sometimes see with greater clarity.
So over time I also expect to write about meditation, neuroscience, psychology, philosophy, philosophy of science, contemplative practice, spirituality, and the question of what it means to build powerful systems wisely.
The through-line is not “AI news.”
The through-line is seeing.
What changed?
What matters?
What is hype?
What is real?
What should builders test?
What does this reveal about minds, machines, incentives, attention, or agency?
That is the kind of writing I want to do here.
Two Layers
Everything in this publication falls into one of two layers.
The seeing layer is the spine: attention, consciousness, contemplative practice, what minds and machines reveal about each other. This is the territory described above, and it stays free. Forever. It is the reason the publication exists, and I want it to reach as many people as possible.
The craft layer is Agentic Engineering Field Notes: the practical material for people who build with AI agents. This is the paid tier.
A Field Note is a specific thing. Each one gives practical advice on how to use AI agents derived from my real-life experience, either in my day job running agentic operations for enterprise clients, or in my home lab, where I build and operate my own agent infrastructure in the open. It covers what I tried that failed, the patterns that actually work, with the actual configs, prompts, and code, and a handful of concrete steps you can run in your own stack.
The “what failed” part is the key. Almost nobody publishes their failures. That is where the real learning happens.
I build my tools in the open, and paid readers will watch new ones take shape here first, in the Field Notes, while they are being built. My last open-source release crossed 285 GitHub stars in its first week; the notes behind work like that are exactly what this tier is for.
Why charge for the craft and not the seeing? Because charging for craft is a fair trade: it has direct value to people who build for a living. What would corrupt this publication is not a paywall. It is paywall-optimized writing. The seeing stays free precisely so I never have to write it with a conversion rate in mind.
Who This Is For
This is for people building with AI who want more than launch-day summaries. Engineers, founders, researchers, product people, agentic engineering practitioners, and technical operators who are trying to understand what new models, papers, evals, and tools mean for actual work.
It is also for contemplative and philosophically minded technologists: people who suspect that the AI story is not only about automation or productivity, but also about cognition, attention, agency, truth-seeking, and human flourishing.
If you want maximal hype, this will probably feel too slow.
If you want detached cynicism, it will probably feel too earnest.
The goal is something else: calm, evidence-backed interpretation for people trying to build and think clearly in a world that’s rapidly evolving.
What To Expect
Here is the honest cadence.
Agentic Engineering Field Notes ship weekly. That is the paid layer’s promise. If a week goes sideways, you will get a shorter pattern snippet rather than silence.
Paying Attention, my brief on model releases, company moves, and AI-builder signals, ships most weeks and stays free.
The seeing essays on human attention, world models, contemplative practice, and philosophy arrive when they are ready. Depth over breadth. The soul’s material is timeless.
Around those anchors you will also see Research Translations that turn papers into practical builder implications, Competitive Maps of the companies and platforms shaping the AI stack, and Reading Bench posts with a few things worth reading, testing, or carrying into the next week.
The Notes will be faster: insights from the essays, a paper quote with the practical implication, a question for builders, a chart or benchmark that deserves interpretation, or a small observation from the workbench.
X is not going away for me. It is still useful for discovery, conversation, and testing ideas quickly.
But I want Substack to become the durable home: the place where the best ideas are easier to find, revisit, forward, disagree with, and build on.
Why Now
AI is moving from chat interfaces into autonomous agentic systems that effect society.
The important questions are becoming less about whether models can produce a good answer in isolation and more about whether they can participate in longer loops:
Can they hold context?
Can they use tools?
Can they recover from mistakes?
Can they verify their work?
Can they coordinate with people and other agents?
Can they improve the system they are part of?
That shift makes the builder side of AI much more interesting.
It also makes the human side more important.
As more cognition gets routed through models, agents, feeds, assistants, recommendation systems, and synthetic text, the quality of our attention starts to matter much more. What we notice, what we trust, what we reward, and what we practice will shape the systems we build and the people we become around them.
That is the reason for this publication.
Not just to follow AI.
To understand what it is doing to work, attention, judgment, and the way we see.
Where to Start
If you are new here, two pieces will tell you whether this room is for you.
Attention is All You Need, but what is Attention? is the seeing layer: the question underneath everything else here.
How I Installed my Soul into an AI Agent is the first Agentic Engineering Field Note: what the craft layer looks like in practice.
Read one of each. If both land, you will know.
An Invitation
If that sounds like your kind of room, subscribe.
Reply and tell me what you are building, what you are watching, what you think I am missing, or which papers, tools, models, practices, and questions deserve a closer look.
I am especially interested in hearing from people working at the edge of AI systems: coding agents, evals, memory, tool use, research translation, human-in-the-loop workflows, contemplative practice, philosophy of science, and the messy boundary between better tools and better attention.
This is the start of the archive I wish I already had:
AI, attention, consciousness, and the practice of living fully.
Attention Heads is a reader-supported publication. The seeing stays free; the craft goes paid. If you want to participate in conversations about AI, Consciousness, Mind and Life, consider becoming a free or paid subscriber.