Deep dives on technology, strategy, and systems.
Current agentic workflows have a dirty secret: they route everything through LLM inference, even the boring, predictable stuff. Need to parse a date? Call the LLM. Format output? C
AI coding agents promise to automate software fixes, but they frequently introduce regressions, breaking tests that previously passed. Benchmarks like SWE-bench emphasize resolut
Source: arxiv.org/pdf/2602.11114v1https://arxiv.org/pdf/2602.11114v1 --- Current agentic workflow systems break when you change domains. New research shows a single-pass gener
Source: arxiv.org/abs/2603.05344https://arxiv.org/abs/2603.05344 --- IDE plugins are dying. The future of AI coding is terminal-native agents, operating where you git, build,
The Problem Interior design has a communication gap problem. Clients struggle to articulate what they want. Designers struggle to explain complex spatial relationships. The res
ArXiv: 2603.12707http://arxiv.org/abs/2603.12707v1, Published 2026-03-13 --- The Problem Running a multimodal LLM vision + language on a single GPU type is leaving money on t
ArXiv: 2603.12091http://arxiv.org/abs/2603.12091v1, Published 2026-03-13 --- The Problem Neural Architecture Search NAS, finding the optimal network design for a task, has hi
ArXiv: 2603.12933http://arxiv.org/abs/2603.12933v1, Published 2026-03-13 --- The Problem Running everything through your best most expensive LLM is a tax you pay on every sin
If you pick a coding agent because its harness scored higher on a leaderboard, you are probably choosing on noise. If you pick it because of what it costs per task, you are choosin
If your multi-agent system fails, you rerun it, and it passes, you have learned almost nothing. That is the uncomfortable result in Repair or Resample? Rethinking Failure Debugging