Weekly dispatch — my thoughts plus the latest AI/ML news that I actually care about. Curated, not exhaustive.

• 2026-03-18
  The Forge #8

  Catch-up edition from Mar 6→Mar 18: agentic coding becomes standard workflow, progress shifts from base-model gains to systems quality, and the platform race broadens across Anthropic, OpenAI,...
• 2026-03-06
  The Forge #7

  Enterprise agent adoption is accelerating, but the real signal is operational friction: context quality, security controls, eval discipline, and infra choices are now deciding who ships durable systems.
• 2026-03-02
  The Forge #6

  Reliability pressure is overtaking raw model hype, agent workflows are becoming operational systems, infrastructure discipline is the new moat, and interpretability tooling is converging into...
• 2026-02-25
  The Forge #5

  Trust overtook raw capability as the bottleneck, governance became a product-surface issue, benchmark credibility fights intensified, and agent/robotics infrastructure kept compounding under the noise.
• 2026-02-23
  The Forge #4

  Agent workflows are moving from demos to operating discipline, infrastructure economics and interconnect constraints are back in focus, and robotics progress is being judged on reliability and...
• 2026-02-20
  The Forge #3

  ClawHub's malware marketplace exposes agent security gaps, Gemini 3.1 Pro doubles ARC-AGI-2 scores, Qwen 3.5 forks from every other frontier architecture, AI drug discovery burns $15B on...
• 2026-02-18
  The Forge #2

  Training costs collapsing ($18 vs $10K for RL), linear attention hitting 19x decoding gains, MIT cracking catastrophic forgetting, Claude Sonnet 4.6 shipping, and Unitree robots doing kung fu for...
• 2026-02-16
  The Forge #1

  Weekly AI research and engineering digest. Interpretability halving hallucinations, model attractor states, synthetic data pipelines, reasoning failure taxonomies, robotics roundup, and agent...