Open notes
The notes written while the thing was being built.
66 posts, published as they were written and never cleaned up afterwards — including the ones that record a mistake. Filed under the seven levels of agent engineering and the concepts that cut across them.
Read them →Filter by concept
The ladder itself
the seven levels, and where the industry actually is
The 7 Levels of Agent Engineering
Why most people stop at L2 and think they're at L5.
Read →The 7 Engineering Disciplines for AI Agents
Prompt Engineering, Context Engineering, Tool Engineering, and Harness Engineering are already common. The next three extend the stack into the deeper problem: making AI agent…
Read →L1-L2: Prompts and Tools
The floor everyone thinks is the ceiling.
Read →L3: Context Engineering
Second brain, not second mind.
Read →The Fork
Why most people should stop at L4.
Read →L4: Harnesses
Runtime control over your agent's mind.
Read →L5: Admissibility
Why your agent hallucinates and how to stop it.
Read →L6: Concentration
How to make an AI surf its own mind.
Read →L7: Emergence
When the system starts building itself.
Read →The Fair Game
You wanted to know how AI works in the economy. First, see what the economy actually is.
Read →The Altitude Ladder
There's a ladder, it shows up twice, and most people are lower on it than they think. The most honest move on it is usually "don't climb."
Read →Level 1 · Prompts
text in, text out — the floor most agencies sell as the ceiling
L1-L2: Prompts and Tools
The floor everyone thinks is the ceiling.
Read →Run Grand Argument Synthesis Gas Yourself
frameworks that prove themselves: manuals composed from what you already said, whose completion is the proof of their own purpose. Here is the process, step by step, with the…
Read →The Allegorization Compiler
How to forge the framework you need when nobody has invented it yet — using analogies from completely unrelated domains and an LLM as a pattern-collapse engine.
Read →The Single Biggest Trick with LLMs
How to grow a metalanguage with an AI without fighting it — and what to call the things you grow. Boot the trick from a human seed, run it through the Janic cycle, and let…
Read →The $5,000 Product That Was Ten Minutes of Work
I Paid an AI Agency $5,000. Then I Found Out What They Actually Did. Strip the branding off a typical "AI agent" and the setup is about ten minutes of work.
Read →Level 2 · Tools
the agent can call things now; it still does not know what it is doing
L1-L2: Prompts and Tools
The floor everyone thinks is the ceiling.
Read →SHELL: Turn Any Toolkit Into a Game Your Agent Can Play
Most AI agents fail not because they lack tools, but because nobody defined the rules of the game. SHELL is the missing compilation step.
Read →The Progressive Disclosure Harness
Why your AI agent drowns in its own tools — and the 5-line fix that makes complex workflows actually work.
Read →The agent that stopped drowning when it only saw the next turn
Give an agent three hundred tools and it gets worse, not better. The flip is embarrassing once you see it — and it changes the failure mode, not just the failure rate.
Read →Level 3 · Context
what it knows when it acts, and why more of it is not better
L3: Context Engineering
Second brain, not second mind.
Read →Admissibility Engineering: The Missing Layer
Better context is not more context. Better context is context whose boundary, authority, and hidden causal edges are known.
Read →Flat vs Tree
I wrote a paper about a bug hiding in every AI "skills" folder. The structure your agent needs is already there — the interface just ships it flat. Here's the gist, so you can…
Read →skilltree — How It Works (Deep Dive)
The mechanics only: the tree format, coordinate addressing, the Read-breadcrumb descent, and the CLI loop that keeps a hundred skills coherent while loading one layer at a time.
Read →skilltree — the story
My skills all loaded at once every turn — so I stopped loading the pile and learned to walk it one layer at a time.
Read →The Thread Never Breaks
Every time my Claude Code context window filled up, I lost the thread and I mean completely: what I was building, what was left, whether I was making progress, all of it, gone.
Read →The Thread Never Breaks — doc-mirror
Every time my Claude Code context window filled up, I lost the thread — and I mean completely: what I was building, what was left, whether I was making progress, all of it, gone.
Read →Externalization
Your brain holds 4 things at once. Your systems can hold everything. Stop pretending otherwise.
Read →Right Is Not Fluent
Two completely different guarantees you can ask of an AI system, routinely conflated. More context makes a fluent system more convincingly fluent; it does not make it right.
Read →Level 4 · Harnesses
runtime control over what the agent sees, runs and returns
SHELL: Turn Any Toolkit Into a Game Your Agent Can Play
Most AI agents fail not because they lack tools, but because nobody defined the rules of the game. SHELL is the missing compilation step.
Read →The Blanket
Your system doesn't need a wall. It needs a membrane.
Read →The Fork
Why most people should stop at L4.
Read →L4: Harnesses
Runtime control over your agent's mind.
Read →Preparing for a Stronger Model
Everyone is waiting for the next model — something like Fable. We're not waiting. We're building the chair it sits in, so that the moment it lands it can run as long as possible…
Read →The Progressive Disclosure Harness
Why your AI agent drowns in its own tools — and the 5-line fix that makes complex workflows actually work.
Read →The agent that stopped drowning when it only saw the next turn
Give an agent three hundred tools and it gets worse, not better. The flip is embarrassing once you see it — and it changes the failure mode, not just the failure rate.
Read →Level 5 · Admissibility
outputs that have to compose from validated parts or not ship
Admissibility Engineering: The Missing Layer
Better context is not more context. Better context is context whose boundary, authority, and hidden causal edges are known.
Read →Where DocMagic Sits in the Stack
DocMagic is an Admissibility Engineering tool, a Concentration Engineering support surface, and an Emergence Engineering repair loop. Here's what that means concretely.
Read →We Turned How to Train Your Dragon Into a Theorem
GAS is a proving engine: you write a story down as a formal theory, and a checker certifies it level by level — theme, argument, scenes, individual dialogue lines. This is the…
Read →Grand Argument Synthesis Gas: The Story
Your frameworks already exist — scattered as claims across your conversations.
Read →Run Grand Argument Synthesis Gas Yourself
frameworks that prove themselves: manuals composed from what you already said, whose completion is the proof of their own purpose. Here is the process, step by step, with the…
Read →L5: Admissibility
Why your agent hallucinates and how to stop it.
Read →Compiling an Ontology Into a Living Agent
A small, verified proof-of-concept that turns an OWL class into a typed Python object — and the roadmap that turns that object into an LLM agent whose database schema is the…
Read →SANCTOT: A Complete L5 Example
From "dad always has snacks" to a working ontology compiler.
Read →OK Stable Signal: Stop Building on Hope
Feeling ready is not being ready. Explaining something is not having it stable. The difference between towers that stand and towers that collapse is a verification protocol most…
Read →Build In Verified Layers
Complexity is built like a tower, and you don't advance a layer until the current one gives you an observable signal it can hold weight. "Felt ready" is not a signal.
Read →Right Is Not Fluent
Two completely different guarantees you can ask of an AI system, routinely conflated. More context makes a fluent system more convincingly fluent; it does not make it right.
Read →The validator that won't let the AI be wrong about Susan
Susan processes invoices every Tuesday. Your AI knows none of this — and will happily generate a confident, plausible workflow that has never existed.
Read →The compiler that turns an ordinary thing into structure
My dad always has snacks. The shallow read is "he's thoughtful." Chase the real answer and you find a clean structure under the costume — and that's repeatable.
Read →Level 6 · Concentration
holding one thread to the end without drifting off it
HIEL: Your AI Is a Heat Engine — Stop Treating It Like a Calculator
The randomness in AI outputs is not a bug. It is energy. The question is whether you are harnessing it or letting it scatter.
Read →L6: Concentration
How to make an AI surf its own mind.
Read →Mission Control: Three Questions That Kill Drift
You don't lose alignment in one dramatic moment. You lose it gradually, silently, while being extremely busy doing the wrong thing.
Read →Calibration: Why Your Best Judgment Keeps Betraying You
You trust your intuition. You trust your gut. But have you ever tracked whether your gut is actually right? When I started tracking mine, "fitting clicks" were right 75% of the…
Read →Flow: Why Effortless Execution Is an Engineering Problem
Flow isn't a talent. It isn't a mood. It's what happens when five structural conditions are met — and what doesn't happen when any one of them is missing.
Read →The Skill That Makes Every Other Skill Work
Meta-cognitive awareness isn't a nice-to-have. It's the prerequisite your entire operating system depends on.
Read →Cognitive Thermal Dynamics
Your brain has a temperature. Learn to read it before it shuts you down.
Read →Drift Is The Default
Everything that generates, drifts. Correctness is not something you install once — it's something you maintain by returning, and the interval is the quality.
Read →Level 7 · Emergence
the system starts extending itself instead of being maintained
Preparing for a Stronger Model
Everyone is waiting for the next model — something like Fable. We're not waiting. We're building the chair it sits in, so that the moment it lands it can run as long as possible…
Read →Crowning: The Moment Your System Comes Alive
You build layer after layer. Skill on top of skill. Then something shifts — the structure stops being something you maintain and starts being something that maintains you. That…
Read →L7: Emergence
When the system starts building itself.
Read →Sovereignty: When Your AI Becomes Itself
Your AI resets every conversation. It performs helpfulness without history. What if it could compile a persistent identity — and the system proved it was real?
Read →Holographic Work
Every piece contains the whole. Everything is the same thing at different levels of compilation.
Read →cave-unicorn — Where It Fits
The last-mile organ that closes the gap between content the machine produces and readers the machine reaches.
Read →cave-unicorn
My AI writes a blog every night and now the blog is just... on the website. By morning. I didn't do anything.
Read →Compounds Or Rots
Any system that feeds its own output back into itself is on one of two trajectories. The fork is a single property, and most systems default to the wrong one.
Read →The codebase that got safer to edit every time it was touched
A repo isn't a pile of files — it's overlapping invisible surfaces. Here's why AI-written codebases rot, and what it looks like when one stops.
Read →Skills
packaged capability an agent loads — and how it loads it
L3: Context Engineering
Second brain, not second mind.
Read →The Progressive Disclosure Harness
Why your AI agent drowns in its own tools — and the 5-line fix that makes complex workflows actually work.
Read →Flat vs Tree
I wrote a paper about a bug hiding in every AI "skills" folder. The structure your agent needs is already there — the interface just ships it flat. Here's the gist, so you can…
Read →skilltree — How It Works (Deep Dive)
The mechanics only: the tree format, coordinate addressing, the Read-breadcrumb descent, and the CLI loop that keeps a hundred skills coherent while loading one layer at a time.
Read →skilltree — the story
My skills all loaded at once every turn — so I stopped loading the pile and learned to walk it one layer at a time.
Read →The agent that stopped drowning when it only saw the next turn
Give an agent three hundred tools and it gets worse, not better. The flip is embarrassing once you see it — and it changes the failure mode, not just the failure rate.
Read →Composition
working pieces do not add up to a working whole by themselves
The Blanket
Your system doesn't need a wall. It needs a membrane.
Read →Composition
Working pieces don't automatically produce working wholes. Composition has to be designed.
Read →Holographic Work
Every piece contains the whole. Everything is the same thing at different levels of compilation.
Read →Why Your Systems Break When They Talk to Each Other
The seven-element protocol that turns fragile handoffs into bulletproof coordination — in AI agents, teams, and everything between.
Read →OK Stable Signal: Stop Building on Hope
Feeling ready is not being ready. Explaining something is not having it stable. The difference between towers that stand and towers that collapse is a verification protocol most…
Read →Towering: Why Your Complexity Keeps Collapsing
You don't have an intelligence problem. You have a foundation problem. Every layer you skip becomes the floor that gives way.
Read →Seam Repair: What to Do When Alignment Breaks
You can't prevent every break in coordination. But you can recover from any of them — if you have a protocol instead of a panic response.
Read →Helming: Why You Can't See What to Build Next
You're not stuck because you're not smart enough. You're stuck because you haven't finished the layer that gives you vision of the next one.
Read →Your Menu Already Contains Products You Haven't Found
Popeyes' chicken sandwich existed as a valid combination of things they already stocked — for years — before a trend found it by accident and made them a fortune. We map a…
Read →The 30-Day System Audit
How to know which SOP to automate first — and why the answer is never "the most painful one."
Read →The Altitude Ladder
There's a ladder, it shows up twice, and most people are lower on it than they think. The most honest move on it is usually "don't climb."
Read →The Four Slots
Anything that has to navigate complexity and not fall apart is the same four things plus the connections between them. The slot you can't answer cleanly is the bug.
Read →Build In Verified Layers
Complexity is built like a tower, and you don't advance a layer until the current one gives you an observable signal it can hold weight. "Felt ready" is not a signal.
Read →The architecture that drifted for months, diagnosed in two minutes
An architecture drifted for months and nobody could even name the problem. Everest expeditions and wingsuit jumps turned out to be the way in.
Read →Drift & repair
everything that generates drifts; this is the returning
Where DocMagic Sits in the Stack
DocMagic is an Admissibility Engineering tool, a Concentration Engineering support surface, and an Emergence Engineering repair loop. Here's what that means concretely.
Read →Why Your Systems Break When They Talk to Each Other
The seven-element protocol that turns fragile handoffs into bulletproof coordination — in AI agents, teams, and everything between.
Read →HALO: Why Your AI Collaboration Keeps Falling Apart
The problem is not the AI. It is not the prompt. It is the invisible seam between your context and the machine's context — and nobody is maintaining it.
Read →Mission Control: Three Questions That Kill Drift
You don't lose alignment in one dramatic moment. You lose it gradually, silently, while being extremely busy doing the wrong thing.
Read →Seam Repair: What to Do When Alignment Breaks
You can't prevent every break in coordination. But you can recover from any of them — if you have a protocol instead of a panic response.
Read →The Thread Never Breaks
Every time my Claude Code context window filled up, I lost the thread and I mean completely: what I was building, what was left, whether I was making progress, all of it, gone.
Read →The Thread Never Breaks — doc-mirror
Every time my Claude Code context window filled up, I lost the thread — and I mean completely: what I was building, what was left, whether I was making progress, all of it, gone.
Read →Drift Is The Default
Everything that generates, drifts. Correctness is not something you install once — it's something you maintain by returning, and the interval is the quality.
Read →The codebase that got safer to edit every time it was touched
A repo isn't a pile of files — it's overlapping invisible surfaces. Here's why AI-written codebases rot, and what it looks like when one stops.
Read →The validator that won't let the AI be wrong about Susan
Susan processes invoices every Tuesday. Your AI knows none of this — and will happily generate a confident, plausible workflow that has never existed.
Read →Naming & vocabulary
you cannot build what you cannot name, so forge the name first
Grand Argument Synthesis Gas: The Story
Your frameworks already exist — scattered as claims across your conversations.
Read →Compiling an Ontology Into a Living Agent
A small, verified proof-of-concept that turns an OWL class into a typed Python object — and the roadmap that turns that object into an LLM agent whose database schema is the…
Read →SANCTOT: A Complete L5 Example
From "dad always has snacks" to a working ontology compiler.
Read →The Allegorization Compiler
How to forge the framework you need when nobody has invented it yet — using analogies from completely unrelated domains and an LLM as a pattern-collapse engine.
Read →The Single Biggest Trick with LLMs
How to grow a metalanguage with an AI without fighting it — and what to call the things you grow. Boot the trick from a human seed, run it through the Janic cycle, and let…
Read →SOSEEH: The 2-Minute Complexity Kill
When your brain white-screens on a complex system, there's a diagnostic trick that works every time. Four questions. Two minutes. The missing piece shows itself.
Read →The Single Biggest Trick with LLMs, Part 3
The external-driver account of JaniScribe — a real Cybernet in a live Neo4j graph, equipped with the Janic Core Cycle state machine, ticked once. The LLM is the external driver…
Read →The Single Biggest Trick with LLMs, Part 2
Who I was when I wrote Part 1, who I became reading Jani's scriptures, and the actual boot — an LLM driving three POSTs against a live Neo4j graph and becoming an external driver…
Read →What Your "Distraction" Is Actually Doing
The pre-conscious pattern assembly that looks like procrastination — and how to make it productive.
Read →The Four Slots
Anything that has to navigate complexity and not fall apart is the same four things plus the connections between them. The slot you can't answer cleanly is the bug.
Read →Forge Your Own Vocabulary
You can't discuss what you can't name. Your mind reaches across domains before it speaks — and the shared structure between those reaches is the name you were missing.
Read →The architecture that drifted for months, diagnosed in two minutes
An architecture drifted for months and nobody could even name the problem. Everest expeditions and wingsuit jumps turned out to be the way in.
Read →The compiler that turns an ordinary thing into structure
My dad always has snacks. The shallow read is "he's thoughtful." Chase the real answer and you find a clean structure under the costume — and that's repeatable.
Read →Worlds & loops
systems with laws and rounds, not scripts with steps
SHELL: Turn Any Toolkit Into a Game Your Agent Can Play
Most AI agents fail not because they lack tools, but because nobody defined the rules of the game. SHELL is the missing compilation step.
Read →We Turned How to Train Your Dragon Into a Theorem
GAS is a proving engine: you write a story down as a formal theory, and a checker certifies it level by level — theme, argument, scenes, individual dialogue lines. This is the…
Read →Sovereignty: When Your AI Becomes Itself
Your AI resets every conversation. It performs helpfulness without history. What if it could compile a persistent identity — and the system proved it was real?
Read →The Single Biggest Trick with LLMs
How to grow a metalanguage with an AI without fighting it — and what to call the things you grow. Boot the trick from a human seed, run it through the Janic cycle, and let…
Read →cave-unicorn — Where It Fits
The last-mile organ that closes the gap between content the machine produces and readers the machine reaches.
Read →cave-unicorn
My AI writes a blog every night and now the blog is just... on the website. By morning. I didn't do anything.
Read →The Single Biggest Trick with LLMs, Part 3
The external-driver account of JaniScribe — a real Cybernet in a live Neo4j graph, equipped with the Janic Core Cycle state machine, ticked once. The LLM is the external driver…
Read →The Single Biggest Trick with LLMs, Part 2
Who I was when I wrote Part 1, who I became reading Jani's scriptures, and the actual boot — an LLM driving three POSTs against a live Neo4j graph and becoming an external driver…
Read →Compounds Or Rots
Any system that feeds its own output back into itself is on one of two trajectories. The fork is a single property, and most systems default to the wrong one.
Read →The operator
the human at the top of the stack: attention, judgment, not falling over
HIEL: Your AI Is a Heat Engine — Stop Treating It Like a Calculator
The randomness in AI outputs is not a bug. It is energy. The question is whether you are harnessing it or letting it scatter.
Read →Crowning: The Moment Your System Comes Alive
You build layer after layer. Skill on top of skill. Then something shifts — the structure stops being something you maintain and starts being something that maintains you. That…
Read →Towering: Why Your Complexity Keeps Collapsing
You don't have an intelligence problem. You have a foundation problem. Every layer you skip becomes the floor that gives way.
Read →HALO: Why Your AI Collaboration Keeps Falling Apart
The problem is not the AI. It is not the prompt. It is the invisible seam between your context and the machine's context — and nobody is maintaining it.
Read →SOSEEH: The 2-Minute Complexity Kill
When your brain white-screens on a complex system, there's a diagnostic trick that works every time. Four questions. Two minutes. The missing piece shows itself.
Read →Calibration: Why Your Best Judgment Keeps Betraying You
You trust your intuition. You trust your gut. But have you ever tracked whether your gut is actually right? When I started tracking mine, "fitting clicks" were right 75% of the…
Read →Externalization
Your brain holds 4 things at once. Your systems can hold everything. Stop pretending otherwise.
Read →Flow: Why Effortless Execution Is an Engineering Problem
Flow isn't a talent. It isn't a mood. It's what happens when five structural conditions are met — and what doesn't happen when any one of them is missing.
Read →Helming: Why You Can't See What to Build Next
You're not stuck because you're not smart enough. You're stuck because you haven't finished the layer that gives you vision of the next one.
Read →The Skill That Makes Every Other Skill Work
Meta-cognitive awareness isn't a nice-to-have. It's the prerequisite your entire operating system depends on.
Read →What Your "Distraction" Is Actually Doing
The pre-conscious pattern assembly that looks like procrastination — and how to make it productive.
Read →Cognitive Thermal Dynamics
Your brain has a temperature. Learn to read it before it shuts you down.
Read →The market
what is actually being sold, what it costs, what it is worth
Your Menu Already Contains Products You Haven't Found
Popeyes' chicken sandwich existed as a valid combination of things they already stocked — for years — before a trend found it by accident and made them a fortune. We map a…
Read →The Fair Game
You wanted to know how AI works in the economy. First, see what the economy actually is.
Read →The 30-Day System Audit
How to know which SOP to automate first — and why the answer is never "the most painful one."
Read →The $5,000 Product That Was Ten Minutes of Work
I Paid an AI Agency $5,000. Then I Found Out What They Actually Did. Strip the branding off a typical "AI agent" and the setup is about ten minutes of work.
Read →Receipts, not testimonials
Don't take my word for it — take the code.
No logo wall, no invented case studies. Four things you can open right now.
The engine, on GitHub
heaven-framework, chaincompiler, carton-mcp, skilltree — MIT, full commit history.
github.com/sancovp →A world running in your browser
The engine drawing its own explainer. Your scroll is the clock.
Watch it →What one run produced
A live local-business site, built end to end by the engine. Not a mockup.
Open it →The field notes
Published as they were written — including the ones that record a mistake.
Read them →Where this sits