November 17, 2025
The LLM structural crisis: solving context decay with the AI Memory Prosthesis
Introducing Simulated Recall via Shallow Indexing (SR-SI), an architectural pattern for reducing context drift in long-running AI workflows.
Sr-si
December 10, 2025
The 200-Prompt Wall
Why long-running AI projects collapse around 200 prompts — and the architectural solution that breaks the limit.
November 10, 2025
AI-assisted development breaks around prompt 200, but it doesn't have to
The 200-prompt wall isn’t a model limitation — it’s a memory architecture problem. SR-SI adds an external memory layer to prevent context collapse.
January 29, 2026
The end of documentation debt: how an AI “memory prosthesis” automates the PRD
PRDs decay on contact with reality. This post outlines an AI-native operating model that turns documentation from writing into evidence-based extraction.
February 14, 2026
What a brick teaches us about AI-driven product development
Spec-first AI workflows manufacture false certainty in unknown territory. Real progress comes from touching constraints first, then documenting what the system proves to be true.
February 16, 2026
I shipped a beta, almost production-grade PWA fast, less than a day
Speed with AI isn’t the hard part. Staying coherent is. Here’s the SR-SI workflow that prevents drift and makes fast builds stay structurally clean.
February 16, 2026
Spec-first is anxiety management
In unknown territory, comprehensive specs don’t reduce risk — they manufacture false certainty. Build small tests, document learning, and iterate with cheap context re-entry.
February 17, 2026
AI can get you 90% of the way to a spec in seconds
AI-generated specs don’t just save time — they frame the problem. That first frame anchors your thinking, narrows the solution space, and can quietly outsource the highest-leverage part of design.
February 18, 2026
SR-SI context savings scale progressively
SR-SI doesn’t just reduce AI re-orientation — it scales with repo complexity. Small repos save 10–20%. Large repos can reach 40–60% when governance stays tight.
February 18, 2026
What if AI could learn to remember?
Memory isn’t storage — it’s reconstruction. SR-SI creates memory-like behavior by using a shallow index as an activation node that triggers architectural re-orientation.
February 19, 2026
Swap the model. The project stays coherent
When the project’s identity lives in an index, not a model, you can switch Claude, Codex, or anything else mid-stream without losing coherence. The mouth changes. The soul stays.
February 19, 2026
The way Einstein's brain worked is how AI should retrieve information
Genius isn’t storing more. It’s retrieving better. SR-SI turns AI retrieval into short, indexed pathways instead of full-context scavenging.
February 20, 2026
AI governance needs identity, not just rules
Most governance focuses on behavior through rulebooks. The deeper shift is building accumulated state and history — giving systems something to be, not just rules to follow.
February 20, 2026
The biggest lie in AI-assisted development
Better specs don’t fix AI projects. Context decay does. SR-SI compresses architectural memory into a shallow index the AI consults and maintains to prevent drift.
February 20, 2026
The moat hiding in plain sight is memory
General AI is stateless by design. The durable moat is structural memory — a system that reconstructs context on demand and compounds coherence over time.
February 21, 2026
I cut the AI's memory and it got smarter
Shrinking a monolithic index into a navigation hub plus scoped sub-indices reduced context overhead and improved coherence. The AI didn’t get smarter — the memory architecture did.
February 22, 2026
SR-SI: The methodology that gives AI persistent memory across any long-running project
SR-SI V2 is live: 106x Token Coherence improvement, near-zero marginal upkeep through agent-operated maintenance, and new findings on functional identity over long-running workflows.
February 23, 2026
I accidentally fixed AI forgetting without embeddings, fine-tuning, or a database
SR-SI fixes long-running AI context decay using a shallow index and a protocol—no embeddings, databases, or fine-tuning required.
February 23, 2026
RAG gives AI a library. SR-SI gives it something closer to a memory
RAG is human-curated retrieval. SR-SI is self-curated reconstruction. That shift is subtle in mechanics, but huge in implications for long-running AI work.
February 23, 2026
The AI isn't an oracle, it's a collaborator
Oracle AI generates synthetic certainty from zero context. Embedded AI collaborates with lived project memory, maintaining continuity across sessions and building from real constraints.
February 24, 2026
The problem with AI isn't intelligence. It's orientation.
Every team I talk to has the same complaint: the outputs are generic. The AI sounds confident but misses what matters. Heavy editing required. Back to square one.
February 25, 2026
How I maintain coherence across 66,000 lines of code without losing the thread
Most AI-augmented development workflows break when they crash into the context wall.
February 26, 2026
SR-SI accidentally solved the documentation problem too
SR-SI forces compact architectural clarity for AI orientation — and that same structure produces always-current human documentation as a byproduct.
February 26, 2026
Why AI outputs feel 60% relevant
When AI lacks your context, it returns the average answer with confident tone. The fix isn’t better prompting — it’s an orientation layer that makes your team’s knowledge findable before execution.
February 27, 2026
Computation is killing collaboration
Most teams treat LLM memory as a compute problem. It’s an architecture problem. SR-SI replaces bloated scaffolds with a simple retrieval prosthesis built on indices, markdown, and Git.
February 27, 2026
Documentation tells you what happened. An index tells AI where to look.
These are not the same artifact and treating them as equivalent is one of the most expensive mistakes in AI-augmented work.
February 27, 2026
I don't run AI training courses. I run AI orientation engagements.
The distinction matters more than it sounds.
February 28, 2026
Building four products in parallel while holding a full-time job — the system, not the hustle
This is not a productivity post. I'm not going to tell you to wake up at 5am.
February 28, 2026
Why I went the opposite direction from spec-first
Spec-first tools aren’t wrong — they just don’t match how everyone thinks. SR-SI supports an architect’s workflow: sketch, test, refine, repeat, without context loss.
March 1, 2026
Why I named the studio Ocarina Systems
An ocarina is a wind instrument you can hold in one hand.
March 3, 2026
I've gotten lazy in the best way
With SR-SI, AI stops being a tool you instruct and becomes a teammate that remembers. I offload file-path and wiring details so I can stay in product thinking and discovery.
March 4, 2026
You don't fix a memory problem by adding more memory
Bigger context windows aren’t memory systems. The fix is structure: a shallow index that points to truth and lets AI re-orient without drowning in history.
March 9, 2026
Encyclopedia vs index: why longer prompts don't fix AI
The problem isn’t how much context you give AI — it’s how findable that context is. Better outputs come from better information architecture, not longer prompts.
March 8, 2026
Don't brief AI with history. Orient it with a map
Good onboarding isn’t comprehensiveness — it’s navigation. SR-SI replaces prompt stuffing with a shallow index that lets AI find the right detail on demand.
March 26, 2026
What SR-SI actually is - and why I stopped explaining it as an AI trick
SR-SI isn’t prompt engineering or RAG. It’s an index architecture that lets models re-orient on demand, preventing coherence drift and turning documentation into a zero-cost byproduct.
March 27, 2026
Gestalt is live. What building it revealed about the problem it solves
Gestalt is a fractal system mapping tool that lets you navigate between strategy and execution without losing context. Built in eight days using SR-SI, it demonstrates how structure—not AI—unlocks speed.
March 30, 2026
Why AI context degrades — and the architectural fix that actually works
AI context drift isn’t a model limitation—it’s an architectural failure. SR-SI replaces brute-force context with indexing, enabling persistent coherence across long-running projects.
March 31, 2026
Protocol: building a fitness OS for coaches who've outgrown their tools
Protocol starts with the athlete but is designed to grow into a B2B2C fitness OS for coaches and boutique studios. The product architecture reflects the business model from day one.
April 6, 2026
AI doesn’t need an encyclopedia. It needs an index.
The problem with most AI workflows isn’t missing information. It’s missing navigation. SR-SI works because it gives AI a compact index, not a bloated encyclopedia.
April 7, 2026
Neon Oracle: why I built an AI tarot interface as a consciousness archiving system
Neon Oracle isn’t just an AI tarot tool—it’s an experiment in using structured sessions as a memory substrate to track patterns in how people think over time.
April 9, 2026
Why I’m building five products at once — the honest answer
Building multiple products in parallel isn’t about hustle or diversification—it’s a response to uncertainty, enabled by infrastructure, and constrained by strict kill criteria.
April 13, 2026
SR-SI in practice: what a context document actually looks like — and what it replaced
A concrete look at how SR-SI works in practice: what the context document contains, how it’s structured, and how it replaces the hidden re-explanation overhead of AI-assisted development.
May 11, 2026
Why your AI context resets every sprint - and what a context architecture actually looks like
Generic AI output usually comes from missing orientation, not weak prompting. Context architecture gives teams a way to preserve decisions, constraints, and product logic across sessions.
May 12, 2026
The memory problem
The first SR-SI lesson was not that AI needed more intelligence. It needed a better way to orient itself before it worked.
May 13, 2026
Memory is retrieval strategy, not storage
Better AI memory does not come from storing more context. It comes from giving the system a disciplined way to reconstruct the right context at the right time.
May 18, 2026
What if your company had a nervous system?
If SR-SI can make AI agents stop forgetting, the larger question is whether the same structure can make an organization hold context across time.
May 25, 2026
What actually deserves to persist?
The same memory discipline that works for codebases and organizations leads to a harder question: what should a life preserve?
May 31, 2026
What Deserves to Persist is ready
The full essay connects earned memory, SR-SI, ephemeral software, organizational memory, and the question of what the digital self should be allowed to preserve.
June 2, 2026
Context decay in AI isn't a model problem. Here's what it actually is.
Context decay usually comes from the way relevant information gets diluted inside long sessions, not from the model suddenly becoming less capable.
June 15, 2026
SR-SI: the methodology for AI that maintains context across a full product build
An accessible explanation of SR-SI as a context architecture for maintaining AI coherence across long-running product builds, teams, and sprint cycles.
June 15, 2026
What coherent AI context looks like across a real sprint - and why most teams never build it
A short sprint-level explanation of why AI context architecture reduces repeated rebriefing and makes product work more coherent over time.
June 29, 2026
The index vs the encyclopedia: why giving AI everything it might need is the wrong move
A short SR-SI methodology essay on why AI context works better as a maintained index than as a bloated encyclopedia of every possible detail.
July 3, 2026
AI isn't proactive. Neither are we.
A context-architecture essay arguing that proactivity is really long-horizon reaction, and that AI workflows start to feel proactive when their input window gets wider.