---
type: "post"
id: "why-ai-assistants-are-slow-and-how-to-9708d9"
slug: "why-ai-assistants-are-slow-and-how-to-9708d9"
title: "Why AI Assistants Are Slow—and How to Make Them Faster"
description: "The bottleneck in consumer agent latency is often orchestration and repeated interface discovery, not model generation speed alone."
retrieval_nugget: "The bottleneck in consumer agent latency is often orchestration and repeated interface discovery, not model generation speed alone."
published_at: "2026-09-25"
updated_at: "2026-09-25"
record_date: "2026-09-25"
date_kind: "published_at"
topics: []
source_urls: ["https://www.cerebras.ai/blog/the-rise-of-slow-personal-assistants"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:why-ai-assistants-are-slow-and-how-to-9708d9"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:038be7794d4ab89f014ca94a8f4900e904d5d0f6cede9bf3d8d868cc335ac25b","reviewed_at":"2026-09-25T04:05:16.675Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":2,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"The bottleneck in consumer agent latency is often orchestration and repeated interface discovery, not model generation speed alone.","observed_facts":[{"text":"Cerebras reports recorded dinner-reservation runs of 4 minutes 36 seconds for Meta Muse, 6 minutes 25 seconds and 57 tool calls for Claude Cowork, and 7 minutes 40 seconds for Grok Bot.","source_urls":["https://www.cerebras.ai/blog/the-rise-of-slow-personal-assistants"]},{"text":"Cerebras says its Qwen 3.8 27B assistant using the Pi harness achieved a 22-second median across two successful attempts and reduced tool calls by more than 80 percent.","source_urls":["https://www.cerebras.ai/blog/the-rise-of-slow-personal-assistants"]},{"text":"Cerebras reports a 6.8-second browser/API segment in its optimized run versus 4 minutes 31 seconds in an earlier Grok trace.","source_urls":["https://www.cerebras.ai/blog/the-rise-of-slow-personal-assistants"]}],"mechanism":"Parallelizing independent availability checks, preserving a reusable site-navigation procedure, and reducing inference pauses remove different components of agent latency.","why_now":"This matters now because personal assistants are shifting from demonstrations toward delegated transactions, where a several-minute wait can erase the convenience of automation.","implications":["The first operational implication is that repeatable web workflows should be encoded as reusable procedures while live facts such as availability and prices remain fresh.","Agent teams should measure end-to-end time across model calls, tool calls, browser waits, and retries rather than optimizing token generation in isolation."],"evidence_boundary":"The evidence boundary is that this is a vendor’s internal, narrowly scoped set of recorded runs, and the reported comparison combines changes to the model, harness, and execution path; it does not identify the causal contribution of any single optimization.","watch_conditions":["The first watch condition is whether independent third-party tests reproduce similar end-to-end gains across multiple websites, models, and agent harnesses.","Watch whether procedural skills remain reliable when websites alter interfaces, authentication requirements, or transaction flows."],"related_records":[],"new_branch_reason":"No prior New Runtime record is supplied; this begins a distinct record for a specific end-to-end assistant-latency benchmark and its stated optimization methods."}
routes: {"html":"https://newruntime.com/posts/why-ai-assistants-are-slow-and-how-to-9708d9/","markdown":"https://newruntime.com/posts/why-ai-assistants-are-slow-and-how-to-9708d9.md","json":"https://newruntime.com/posts/why-ai-assistants-are-slow-and-how-to-9708d9.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

The bottleneck in consumer agent latency is often orchestration and repeated interface discovery, not model generation speed alone.

The bottleneck in consumer agent latency is often orchestration and repeated interface discovery, not model generation speed alone.

Cerebras recorded several assistants attempting the same dinner-reservation request: Meta Muse took 4 minutes 36 seconds, Claude Cowork took 6 minutes 25 seconds and made 57 tool calls, and Grok Bot took 7 minutes 40 seconds. In its own test configuration, Cerebras used Qwen 3.8 27B with the Pi agent harness and reported a 22-second median from two successful runs. It also says the optimized test reduced tool calls by more than 80 percent.

Parallelizing independent availability checks, preserving a reusable site-navigation procedure, and reducing inference pauses remove different components of agent latency. The company’s browser/API comparison—6.8 seconds in its optimized run versus 4 minutes 31 seconds in an earlier Grok trace—illustrates the gain from changing the whole execution path rather than waiting on checks sequentially.

This matters now because personal assistants are shifting from demonstrations toward delegated transactions, where a several-minute wait can erase the convenience of automation. Teams building agents should treat tool-call topology, procedural memory, and inference latency as one performance budget. The first operational implication is that repeatable web workflows should be encoded as reusable procedures while live facts such as availability and prices remain fresh.

The evidence boundary is that this is a vendor’s internal, narrowly scoped set of recorded runs, and the reported comparison combines changes to the model, harness, and execution path; it does not identify the causal contribution of any single optimization. The first watch condition is whether independent third-party tests reproduce similar end-to-end gains across multiple websites, models, and agent harnesses.

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: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- 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)