Glassbox: Making AI Agent Context Observable
How Glassbox finds provably dead context in coding-agent sessions and removes it with a validated, lossless fork instead of a lossy summary.
Pi Harness Is a Gamechanger for Coding Agents
Why Pi's minimal, extensible terminal harness beats heavier agent workflows by making the model, tools, session, and product surface composable.
Rote: Memoizing Agent Trajectories Instead of Re-Deriving Them
A technical look at Rote's tool-boundary recorder, typed playbook DAG, assertion-gated executor, environment fingerprints, and planned self-healing repair loop.
I built the missing layer on top of Crustdata's API
The gap I kept hitting: every existing way to plug Crustdata into an agent is pull-only —> search, enrich, fetch. Nothing turns the real-time data into a *trigger*..
Prompting is Not Just Plain English
How attention dilution, vague quantifiers, and structural noise silently degrade your LLM outputs — and what the research says about fixing it.
Graphify × SWAP Integration
Graphify gives SWAP semantic priority, graph-grounded arbitration, and pre-claim coordination across the repo.
Letta and the Context Window Problem Nobody Is Solving Correctly
Letta treats the context window like L1 cache, not memory, and builds an actual memory hierarchy around that idea.