Determinism Injection
Large prompts drift when they do deterministic work themselves. Determinism Injection hands that work to tools instead so that the prompt stays the orchestrator, and binaries handle the lookups, state transitions, and writes.
Large prompts drift when they do deterministic work themselves. Determinism Injection hands that work to tools instead so that the prompt stays the orchestrator, and binaries handle the lookups, state transitions, and writes.
Introducing Retrofit, a CLI that records database schema evolution as logical, identity-preserving operators instead of hand-written DDL migrations.
Notes on the current LLM development toolchain I am using in June.
Evidence-based iteration on the RDR specification-first workflow: indexing research and Claude Code sessions in Arcaneum to mine and formalize the patterns I was already applying by hand.
17 RDR postmortems, 208 drift instances, one meta-lesson: specify interfaces not implementations, verify APIs against source code, and plan for the full scope including Day 2 ops.
Building Arcaneum using specification-first LLM development with RDRs, writing dozens of specs before a single line of Python code.
Introducing RDRs - specification documents that evolve during planning to prevent purpose drift and wasted implementation effort.
Fronting a private AppRunner instance with API Gateway, by way of the L1 constructs and VPC plumbing the CDK does not expose.
Exporting CloudWatch logs via the CreateExportTask.
Names of things can collide with other things.
Faster CloudFormation stack updates, via the Java CDK, by using ordered Maps.
Cascading 3.0 proves we can abstract away the underlying platform by providing additional platforms users can execute against, not just MapReduce. Apache Tez is the first such addition. Tez is the latest parallel computation fabric from the Apache group. It builds on Hadoop, keeping the Hadoop File System for instance, but uses a model that removes inefficiencies that are baked into the MapReduce model.