Human-Assisted AI
What AI Is, How It Fails, and Why Your Engineering Discipline Is Your Best Defense.
Read this if the work keeps going wrong in ways that feel familiar but still do not have names.
Read the publication pageThe Meridian Model · Four Publications
AI-assisted work keeps failing in recognizable patterns.
The Meridian Model is a body of work comprising Human-Assisted AI, The Confluent Method, The Halocline, and The Discipline of Dependable Software. Together they name the failure patterns of AI-assisted work, define the method that responds to them, mark the boundary where that method stops, and anchor the whole system in the engineering philosophy underneath.
The publications
Four works. One system.
Human-Assisted AI is the gateway. The Confluent Method is the practical path. The Halocline is the boundary branch. The Discipline of Dependable Software is the foundation underneath the whole model.
The Confluent Method
The Structured Process for AI-Assisted Development.
Read this when the question becomes: what do I actually do on Tuesday morning?
Read the publication pageThe Halocline
The Invisible Boundary Between Two Kinds of AI Work.
Read this when the question becomes: what about agentic AI?
Read the publication pageThe Halocline is a branch from the method boundary — important, but not the gateway.
Choose your entry point
What problem brought you here?
You can read the works in dependency order if you want the philosophy first. Most people arrive because one specific problem is already happening to them. Pick the one that sounds like yours.
I keep getting plausible but wrong output, and I can't always explain why.
Human-Assisted AIThe failure patterns, named.
Read this one →
I need a working method for creative AI-assisted development.
The Confluent MethodThe practice that handles the failures.
Read this one →
I'm trying to classify where agentic AI fits — and where it doesn't.
The HaloclineThe boundary, and why it matters.
Read this one →
I want the engineering philosophy that makes all of this possible.
The Discipline of Dependable SoftwareThe foundation underneath.
Read this one →
Supporting asset · From Human-Assisted AI
The Failure Mode Catalog.
Eight patterns most readers have already lived through.
Silent Divergence. Compressed Context Corruption. The Spiral. The Tool Trap. Collateral Mutation. Confidence Poisoning. The Phantom Agreement. Test Laundering.
Once the patterns have names, teams stop treating them as isolated surprises. That is why Human-Assisted AI works as the gateway into the rest of the model — recognition is what pulls a practitioner into the method.
See the catalog in contextHow the works relate
One system. Two ways to read it.
Most readers enter through recognition. Readers who want the philosophy first can work from the foundation upward. Both paths lead to the same place.
Foundation
The Discipline of Dependable Software. The engineering philosophy beneath every other work — understandable, traceable, controllable systems.
Recognition layer
Human-Assisted AI. The failure surfaces, the human role, and the reason discipline matters before any method is applied.
Practice layer
The Confluent Method. The actionable method for Creative AI Domain work. Small steps, verified output, human-confirmed completion.
Boundary branch
The Halocline. The consequence of discovering that the method stops at the agentic / OAID boundary — and what to do with that.
Published & citable
All four works were first published in April 2026 and archived on Zenodo with persistent DOI identifiers for durable citation.
Open repository
The open GitHub repository holds canonical PDFs, companion cards, proposal materials, and change history.
Author & publisher
Written by Phil Russo. Published through Riverbend Consulting Group. Licensed CC BY 4.0 for reuse with attribution.
Read the works. Use the catalog. Apply the method.
The works are free, citable, and open. Start with recognition, work outward toward method and scope. Or start at the foundation and work up. Both paths lead to the same place.