SemiAnalysis: Persistent Agents Take 45% of AI Spend, Perplexity Beats Claude and OpenAI on Experience

SemiAnalysis analyst Dylan Patel argues that "persistent agents are the next unlock for AI usage," sharing his team's internal AI spend dashboard as first-hand evidence: as of the week ending November 8, 45% of the team's AI spending went to persistent agents. Based on this, he offered sharp takes on the current agent product landscape: Anthropic's Claude Tags offers a poor experience, and OpenAI's Slack agent also falls short.

Confirmed

• Data from Dylan Patel's team: as of the week ending November 8, 45% of AI spend went to persistent agents.
• His team's use cases are driven mainly by Anthropic models, but he believes Anthropic "builds poor products and isn't very receptive to product feedback," while OpenAI has product weaknesses of its own—meaning the two vendors with the strongest models actually lag in agent product experience.
• Why he favors Perplexity: though not the category's inventor, as a fast follower it can ship any requested feature in under 24 hours; he also shared the anecdote that the name "Computer" was suggested by them.
• Dylan Field reposted the same view, echoing the judgment that "persistent agents are the next growth driver for AI usage, with 45% of spend going to persistent agents."

Why it matters

• This is first-hand spending data from a top analysis firm, showing persistent agents evolving from concept into a real and sizable usage category—potentially signaling a new direction in AI consumption patterns.
• The "strongest models ≠ strongest products" assessment points to competitive pressure on Anthropic and OpenAI at the agent product layer, while fast-follower model application vendors like Perplexity are seizing the high ground in user experience.

2026-11-12 ~ 2026-11-12 · 6 related posts

• [source] SemiAnalysis: Persistent agents now eat 45% of our AI spend, and Perplexity's harness beats Claude Tags — dylan522p · 2026-11-12
• [source] Anthropic models power the work, but Anthropic 'sucks at product', says SemiAnalysis — dylan522p · 2026-11-12
• 'They even got the name from us': Perplexity ships feature requests in under 24 hours — dylan522p · 2026-11-12
• Heavy User: 45% of Our AI Spend Now Goes to Persistent Agents, Perplexity Leads — zealcaiden · 2026-11-12
• 45% of our AI spend goes to persistent agents, says Dylan Field, praising Perplexity — AravSrinivas · 2026-11-12
• Persistent agents eat 45% of one team's AI spend; Perplexity's harness beats Claude and OpenAI — petrusenko_max · 2026-11-12