Agentic test processes, LLM benchmarks, and other notes on agentic coding fr
THE SO WHAT
Dan Luu’s analysis of agentic coding and LLM variance underlines a hard constraint—without disciplined test processes and benchmarks, “AI wrote it” is un-auditable. Engineering leaders experimenting with agentic workflows should invest more in harnesses and evals than in yet another model swap.
READ THE SOURCE
MORE FROM THE WIRE
Applied AI‘We cannot choose to become idiots’: a Brown professor’s proof of mass AI cheating
A midterm average of 96 dropping to 48 when exams move in-person is a stark measure of how deeply AI has permeated unsupervised work. Enterprises relying on take-home tests — coding challenges, case studies, written assessments — should assume AI assistance is the default and redesign evaluation around supervised, live, or AI-explicit formats.
Applied AIThe Future Is Always Listening: OpenAI Says Its New Voice Assistant Is ‘One Step Closer to a Truly Accessible AGI’
A more natural, always-listening voice assistant from OpenAI pushes AI deeper into continuous, ambient interaction — not just query/response. Product teams should be planning for voice as a primary interface in workflows where typing is currently the bottleneck, and revisiting privacy and consent models accordingly.
Applied AISpaceX's Grok 4.5 launches at half the price of rivals — here's why that could rattle Anthropic and OpenAI
A coding- and agent-focused Grok 4.5 at roughly half the price of rivals, backed by a $60 billion Cursor acquisition, turns AI coding into a margin-compression arena. For engineering leaders, this accelerates the move from ‘which model’ to ‘what’s the all-in cost per developer workflow’ — expect vendors to compete on integrated agents, not just tokens.
Applied AIAI Spending Raises Questions on Diverging Chip & Hyperscaler Cycles
If chip makers are ramping while hyperscalers start to question AI capex intensity, we’re entering a phase where supply and demand cycles may decouple. Enterprise buyers should watch for a window of better pricing and capacity — and avoid locking into long, inflexible commitments right as hyperscalers reassess spend.