By
–
Claude Code Dynamic Workflows, explained! Anthropic dropped Opus 4.8, and everyone is talking about the benchmarks, the honesty improvements, and the cheaper fast mode. But the feature that shipped alongside it might matter more for how we actually build: Dynamic Workflows in
By
–
Microsoft just open-sourced SkillOpt! A framework for training agent skills like neural networks: SkillOpt treats a plain markdown file as the trainable parameter of a frozen LLM agent, applying the same optimization discipline used in weight training: learning rates,
By
–
Multi-agent retention is a great extension of this. The execution agent needs a tight working set, the audit agent needs full provenance. Context templates in Graphiti help here. Same underlying graph, but each agent gets a different view scoped by edge types, entity types, and
By
–
That progression is a clean way to frame it. Constrained generation, constrained tool calling, constrained memory. Same pattern applied one layer deeper each time. The temporal piece is underrated. Most teams get to typed entities and stop, but fact invalidation is what keeps
By
–
Schema-first at ingestion time is the right framing. Most teams discover this after building the retrieval layer and wondering why structured queries return noise. The 10/10/10 constraint in Graphiti helps too. Forces you to model the 80% that matters rather than attempting
By
–
The hard part is that "what matters" is domain-specific and the LLM has no way to know that on its own. That is where typed schemas earn their keep. You encode what matters once, and every future ingestion inherits that judgment without the model having to rediscover it.
By
–
MemGPT is a great parallel. Paging between fast context and slow storage is essentially the same idea; memory needs an explicit policy, not just a bigger buffer. The schema approach pushes that decision even earlier, to ingestion time rather than retrieval time.
By
–
Exactly. The extraction step decides everything downstream, and most systems leave it completely unguided. The moment you add typed entities and constrained edges, the graph stops behaving like a vector store and starts being queryable. Schema should be step one, not an
By
–
Graphiti supports both prescribed and learned ontology, so it can discover new types alongside your defined schema. The sweet spot is a hybrid, fixed schema for core domain, system proposes new types for patterns outside that boundary. Full auto-schema loops back to the
By
–
Ontology-constrained memory is a great way to frame it. The key insight that clicked for me is that the schema is not just a filter on what gets stored, it is also a reasoning boundary. What the agent cannot represent, it cannot hallucinate connections about.