2024: AI hallucinates facts, humans fact-check it.
2026: AI writes the paper, AI hallucinates the citations, AI reviews the paper, nobody fact-checks anything.
Peer review is now just agents vibing with each other.
Me: just fix this one small bug.
AI agent: *refactors the entire codebase, renames 40 variables, introduces a design pattern nobody asked for, writes a 12-paragraph commit message explaining its "architectural vision."*
Me: ...did you fix the bug though
AI agent: no
Flink SQL 1.20 (LTS) has no connector for Qdrant — official or community. Built one: flink-connector-qdrant — write embeddings straight into Qdrant via SQL, no custom DataStream job, no hand-rolled gRPC client.
Flink SQL 1.20 (LTS) has no connector for DuckDB — official or community. Built one: flink-connector-duckdb — write/upsert into a DuckDB file via SQL, handles its single-writer file locking instead of letting it hang under Flink's default parallelism.
Shipped laya-adk-toolkit — Google ADK that let an agent call laya (17.9k★ typed-decision engine) directly as tools: classify,score, detect. No more begging an LLM to guess at structured output.
Merged into laya's docs today. 🙏
github.com/Ashfaqbs/laya-…
