The a16z Show · Sep 29, 2026 · 50m · 30m saved
With David Pawlan, Creator of Assistant Benchmark · Anish Acharya, General Partner at Andreessen Horowitz
Of the 122 personal AI assistants David Pawlan has cataloged, 65 charge their users and 13 are free. The two that lead the market, Meta's Muse and Instinct, are among the free ones.
Most of the attention on these agents goes to viral tricks such as checking a user into a flight. Pawlan said the winner will be the agent people barely notice, one that saves them money in the background, and Anish Acharya said the harder problem is cost: by a16z's estimate an ambitious agent costs something like $20 per user per day to run.
"I think there is massive defensibility around proactivity."
David Pawlan, creator of Assistant Benchmark, a site that runs the same everyday prompts through AI assistants and scores the results, talked with Anish Acharya, General Partner at Andreessen Horowitz, on The a16z Show. Pawlan said the site drew more than 100,000 visitors in the 16 days after it launched. The conversation was recorded on September 24.
The full interview is covered here so you can skip it. 50 minutes of audio, 20 minutes of reading.
Here are the 18 arguments that matter.
Key Takeaways
65 of the 122 agents Pawlan has cataloged charge money, while Muse and Instinct give theirs away
Acharya put the cost of an ambitious agent at something like $20 per user per day
Most consumers do not care about being 10% more efficient, Pawlan said, so the useful agents will work out of sight
The best early use cases save money, from HSA reimbursements to flight credits to a sprinkler that cut a water bill by 50%
Personality is easy to copy; proactivity is not, and one overstep can cost an agent the user's trust
Today's consumer agents are mostly OpenClaw with the setup done for you, Pawlan said
Acharya's narrow-startup thesis: a $100M run-rate business built on tens of thousands of users paying $200 to $300 a month
Amazon earns from ads that agents never see, which Pawlan said explains why it blocked Muse while Shopify let it in
Acharya would rather see an agent worth $1,000 a month than one whose main appeal is a subsidy
1. The 122-Agent Benchmark
Acharya opened by listing three moments when part of the technology world had "seen God": ChatGPT in late 2022 and 2023, coding agents such as Claude Code and Codex, and most recently personal agents such as Instinct and Muse. Pawlan said the space had exploded over the past four weeks. He was an early user of Poke, one of the first consumer AI agents, and started using it three days after it launched. He said OpenClaw's release showed that something was coming, and that Instinct then set off a run of launches, including Muse and smaller agents.
The benchmark runs identical prompts through every assistant
Pawlan said the site gives each assistant the same one-shot request, such as booking a flight to Chicago or finding a vegetarian restaurant within five blocks, and scores the outcome on 16 dimensions, including whether it asks follow-up questions and how fast it replies.
"Assistant Benchmark is a site that essentially compares all these AI assistants on a use case basis." — David Pawlan
He said it started as his own market research and is meant as a consumer-facing guide rather than a technical benchmark.
Most of the market is chasing the general-purpose assistant
Pawlan split the field into consumer-facing generalist agents, such as Instinct and Muse, with, he believes, 64 in the general category on his site; specialists for travel or email; and business workflow assistants that act more like an executive assistant.
"So far to my knowledge there are 122 different ones." — David Pawlan
He said nearly everyone is currently building the generalist.
