Signals — AI Prompt Quality & Task Engineering

Articles and insights on AI prompt quality.

March 12, 2026

A GDPVal task inspected: what a knowledge-work brief looks like when the model has to ship documents

ChatGPT with GPT-5.4 Thinking took ~26 minutes and produced a master schedule plus two applicant itinerary documents from a GDPVal task. The result is already useful work — and the brief quality drove it.

Read →

February 27, 2026

Prompt Optimizers Are Obscuring Your Intent

OpenAI's Prompt Builder turned 15 words into 350 — most of it noise. The real reliability drivers are schema and case specificity. Everything else is filler between you and the result.

Read →

February 20, 2026

GDPVal benchmarks AI against knowledge work worth $3 Trillion in U.S. annual wages

OpenAI's GDPVal tests frontier models on 1,320 real knowledge-worker tasks across 9 sectors. The biggest performance levers: more context, intermediate steps, and prompt optimization.

Read →

February 12, 2026

Put the discipline in before you hit enter. The AI won't have your back just yet.

A paper (GEPA) shows systematic prompt refinement can outperform reinforcement learning — without any model retraining. The quality gap may not be a model problem, but a briefing problem.

Read →