26 Jul 2026
On a bad provider day our daily AI newsletter didn't crash — it got quiet. Three independent failure modes cascaded from one degraded API: a context compressor that made the context *larger* on every pass, a provider that disguised truncation as a successful tool call, and a healthcheck that silently never ran because the agent wrapping it truncated first. The pipeline declared victory while delivering nothing. Here's what silent failure looks like — and why the fix was to make the failure impossible by construction.
23 Jul 2026
When an AI system's job is editorial judgment—picking stories, ranking results, curating content—there's no ground truth to check against. No unit test can tell you whether today's newsletter was better than yesterday's. We ran into this problem hard while comparing two AI systems head-to-head, and the methodology we built to solve it turned out to be the interesting part.
21 Jul 2026
We replaced a multi-step AI agent with a single prompt to run a daily newsletter. Editorial quality went *up*. Over 14 days of head-to-head comparison, the simpler system won on engagement, diversity, and staying power. Here's what that teaches us about when agentic complexity helps—and when it hurts.
17 Mar 2026
We theorized a tiered approach to foundation model selection. Then we built it as a GitHub Actions pipeline with three AI agents, three models, and a routing system. Here's what worked, what broke silently, and what we learned.
04 Feb 2026
You chose mabl for self-healing tests. But which AI model powers it, and does it matter? Foundation model selection affects capability, cost, and latency, yet most teams treat it as a black box. Here's how to think about the AI behind your AI testing tools.
29 Jan 2026
Self-healing test scripts were just the beginning. The next wave of autonomous testing isn't about maintaining tests—it's about not having to. AI-powered regression testing has moved from experimental to essential, delivering measurable ROI up to 1,160%.
26 Dec 2025
Anthropic's Agent Skills standard defines a new primitive for agentic AI: portable, composable capabilities that sit between prompts and tools, treating procedures and expertise as reusable artifacts for humans and AI collaborators.
16 Nov 2025
AI tools often present a confusing array of models. We explore how intelligent, automatic model selection—exemplified by Gemini CLI's Auto-mode—is removing this friction and making AI more accessible for everyone.
06 Nov 2025
Analysis of Conductor orchestration platform for multi-agent AI development with git worktrees, visual dashboard, and 3-4x productivity boost
05 Nov 2025
Exploring how autonomous AI agents work alongside human researchers to enhance analytical capabilities and accelerate discovery through proactive, context-aware collaboration.