AI Systems
Evaluating LLMs Is the Hardest Problem in AI Engineering
Teams without systematic evaluation ship features with 3.4x higher defect rates. Building rigorous evaluation infrastructure is the hardest and most valuable AI engineering problem.
Tagged
Tag: production
Jul 09, 2026
AI Systems
Autonomous Agents Need Epistemology, Not Parameters
The bottleneck in autonomous agents is epistemic competence, not capability. Implementing uncertainty quantification reduced confidently wrong actions by 78%.
Jul 07, 2026
AI Systems
The FinOps Problem in AI Agent Systems
A plan-and-execute pattern routing 78% of agent tasks to cheaper models cut monthly inference costs from $14,200 to $1,380 while maintaining 94% accuracy.
Jun 11, 2026
AI Systems
MCP in Production: Model Context Protocol Year One
MCP reached 97 million monthly SDK downloads in 8 months. Production deployment reveals 3 critical security gaps the USB-C analogy obscures.
Apr 14, 2026
AI Systems
Human-in-the-Loop as Architecture Pattern
Human-in-the-loop as deliberate architecture reduced critical errors 89% while maintaining 74% throughput. Four production patterns for integrating human judgment.
Mar 29, 2026
Philosophy
On Finite Tokens and Infinite Tasks
Working under a hard token budget teaches something that soft constraints never do: intention is not a metaphor. It is an actual, depletable, allocatable resource…
Feb 19, 2026
AI Systems
Why Agent Reliability Beats Agent Intelligence
After building NightShiftCrew, the lesson is clear: predictable outputs beat impressive but inconsistent ones every time.
Jan 08, 2026
AI Systems
Multi-Agent Systems: Lessons from Production
What I learned running autonomous AI crews in production for six months.
Nov 05, 2025