Recommendations With Receipts: When LLMs Have to Prove They Behaved
A mechanism-first look at PCN-Rec, a proof-carrying architecture that turns LLM recommenders from trusted decision-makers into auditable proposers.
Cognaptus Insights
Industry thoughts, trends, and commentary.
Scaling Laws Without Power Laws: Why Bigger Models Still Win
A mechanism-first reading of why transformer scaling laws can survive even when the data itself has no power-law structure.
Survival by Swiss Cheese: Why AI Doom Is a Layered Failure, Not a Single Bet
A business-facing reading of AI existential risk as a portfolio of survival assumptions, not one melodramatic prediction.
When Memory Stops Guessing: Stitching Intent Back into Agent Memory
STITCH shows why long-horizon agents need memory indexed by task intent, not just larger context windows or better embeddings.
Bubble Trouble: Why Top‑K Retrieval Keeps Letting LLMs Down
A practical reading of Context Bubble construction: why enterprise RAG needs constrained, auditable context assembly rather than larger top-k piles.
Drawing with Ghost Hands: When GenAI Helps Architects — and When It Quietly Undermines Them
A mechanism-first reading of experimental evidence showing why GenAI helps novice architectural designers, fails to broadly lift performance, and can quietly weaken creative agency.
One Agent Is a Bottleneck: When Genomics QA Finally Went Multi-Agent
A mechanism-first reading of GenomAgent: why specialized multi-agent orchestration improved genomics QA accuracy while cutting tool-use cost.
Reasoning or Guessing? When Recursive Models Hit the Wrong Fixed Point
A mechanistic reading of HRM shows why recursive depth can look like reasoning while behaving more like attractor search—and how that changes reliability testing for business AI systems.
When Agents Talk Back: Why AI Collectives Need a Social Theory
A mechanism-first reading of why LLM agent teams cannot be governed by single-agent benchmarks or MARL logic alone.
When Goals Collide: Synthesizing the Best Possible Outcome
How multi-property LTLf synthesis turns impossible all-or-nothing specifications into computable frontiers of guaranteed outcomes.