How long can your skills be before your agent forgets what you told it?

Official Schedule Context

Official Description

A year ago, frontier models lost the thread somewhere around 200 simultaneous instructions, so

skills files had to stay short and lean on sub-skills and subagents. We re-ran IFScale on the 2026

frontier and found the ceiling has moved by an order of magnitude: closer to 2,000 instructions, up

to 5,000 on the strongest models. The more interesting story is how models fail at the new frontier:

DeepSeek quietly drops instructions, Opus refuses outright when innocuous words trip a safety

classifier, Gemini burns its whole budget on reasoning and emits nothing, and GPT-5.5 stops to tell

you your request was unreasonable. The capacity problem is largely solved; verification is wide

open. We'll show the data, the failure modes, and what it costs to find out. You’ll come out with

hard data on the ceiling for complex instructions to LLMs

Related YouTube Video

Ship Real Agents: Hands-On Evals for Agentic Applications — Laurie Voss, Arize (speaker-match related prior/adjacent AI Engineer video; captions: English auto-captions).

Transcript Status

Related video transcript availability: English auto-captions. Treat this as supporting context, not a recording of this exact scheduled session unless later confirmed. Cached at raw/sources/youtube-transcripts/Xfl50508LZM.txt (22,591 words).

People

Notes

Supporting Slides

Slide Evidence