AI Trends

Trending story

Philipp Schmid shares an experiment on deleting 90% of the runtime framework

1 reports1 story updatesUpdated Nov 16 14:14

Understand this story

AI overview

At the AI Engineer Code Summit 2026, Google DeepMind engineer Philipp Schmid shared an experiment titled "I deleted 90% of my harness (And It Got Better)." He reduced the coding agent's tools from 10 to 20 specialized tools down to just a single Bash, and cut the system prompt from about 12,400 tokens to 574. Using Gemini 4 Argon, he ran four runtime frameworks on more than 350 of his own tasks; the score gap was within 6 points and token usage differed by 39%, but the cost per item ranged from $2.33 to $5.33—using only Bash scored highest yet was the most expensive. He also compared Gemini 3.5 Flash with Gemini 4 Carbon, showing that the stronger the model, the better it can merge multiple steps into a single call.

AI-generated from public reports

Latest development · Report

Google DeepMind engineer Philipp Schmid says it works better after deleting 90% of the harness

Developments

1 confirmed developments

  1. Philipp Schmid shares an experiment on deleting 90% of the runtime framework

    Source:Shao Meng

Reports and discussion timeline

Browse reports and discussion by time. Their evidence levels remain distinct.

Nov 16

  1. Shao Meng

    Google DeepMind engineer Philipp Schmid says it works better after deleting 90% of the harness

    Google DeepMind engineer Philipp Schmid shared "I deleted 90% of my harness (And It Got Better)" at the AI Engineer Code Summit 2026, streamlining the coding agent's tools from 10 to 20 specialized tools down to just one Bash, and reducing the system prompt from about 12,400 tokens to 574. He used Gemini 4 Argon to run four runtime frameworks on more than 350 of his own tasks; the score gap was within 6 points and tokens differed by 39%, but the cost per item ranged from $2.33 to $5.33, with Bash-only scoring the highest yet costing the most. He also compared Gemini 3.5 Flash with Gemini 4 Carbon, showing that the stronger the model, the better it can merge multiple steps into a single call.

Related stories

  1. Related storiesAnthropic Publicly Submits S-1 for Proposed Nasdaq IPO
  2. Related storiesMeta releases open-weight model Muse Spark Open
  3. Related storiesLauren Tan announces that her AIE talk is now online
  4. Related storiesAddy Osmani writes about the Claude Code team's verification methods