---
type: "raw_signal"
id: "why-ai-assistants-are-slow-and-how-to-d1ccff"
slug: "why-ai-assistants-are-slow-and-how-to-d1ccff"
title: "Why AI Assistants Are Slow—and How to Make Them Faster"
description: "Cerebras published recorded benchmarks of AI assistants completing the same dinner-reservation request."
retrieval_nugget: "Cerebras published recorded benchmarks of AI assistants completing the same dinner-reservation request."
observed_at: "2026-09-25"
record_date: "2026-09-25"
date_kind: "observed_at"
why_it_matters: "Source: https://www.cerebras.ai/blog/the-rise-of-slow-personal-assistants"
novelty: "notable"
verification_level: "source-linked"
signal_type: "research"
source_platform: "cerebras.ai"
topics: []
entities: []
related_patterns: []
source_url: "https://www.cerebras.ai/blog/the-rise-of-slow-personal-assistants"
source_urls: ["https://www.cerebras.ai/blog/the-rise-of-slow-personal-assistants"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "signal:why-ai-assistants-are-slow-and-how-to-d1ccff"
status: "published"
visuals: []
routes: {"html":"https://newruntime.com/signals/why-ai-assistants-are-slow-and-how-to-d1ccff/","markdown":"https://newruntime.com/signals/why-ai-assistants-are-slow-and-how-to-d1ccff.md","json":"https://newruntime.com/signals/why-ai-assistants-are-slow-and-how-to-d1ccff.json"}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
---

# Why AI Assistants Are Slow—and How to Make Them Faster

## Retrieval answer

Cerebras published recorded benchmarks of AI assistants completing the same dinner-reservation request.

Cerebras published recorded benchmarks of AI assistants completing the same dinner-reservation request. Its examples report Meta Muse at 4 minutes 36 seconds, Claude Cowork at 6 minutes 25 seconds with 57 tool calls, and Grok Bot at 7 minutes 40 seconds. Cerebras’ own Qwen 3.8 27B assistant, using the Pi harness, recorded a 22-second median across two successful attempts.

The company attributes the optimized path to parallel availability checks, a saved website-navigation procedure, and faster inference. It reports that its browser/API segment took 6.8 seconds versus 4 minutes 31 seconds in an earlier Grok trace, while tool calls fell by more than 80 percent. Cerebras cautions that this bundles model, harness, and execution-path changes rather than isolating any one intervention.

Source: https://www.cerebras.ai/blog/the-rise-of-slow-personal-assistants

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- collection: [Raw Signals](https://newruntime.com/signals.json): Open every source-linked raw signal.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)