Hayden Field: Meta's Muse, OpenAI's Dots push agents to millions

Field traces the current wave of consumer agents back to OpenClaw, a homebrew tool built by one man, Peter Steinberger, over a weekend. Its technical approach, a harness wrapped around a model with a browser, became the default winning model. OpenAI hired Steinberger, and Meta says it built Muse from scratch but was inspired by OpenClaw.

Meta's advantage, Field argues, is product design and distribution rather than frontier capability. She says Meta still does not have a frontier model, but it is better at making consumer products, does not pay for advertising on its own platforms, and pushes Muse with one-tap downloading and pop-ups across apps like Instagram; the goal is to make Muse the default for the general public. OpenAI is going the other way, marketing Dots to enterprises and knowledge workers and talking up privacy to set it apart from Muse, which Field links to OpenAI needing money ahead of its IPO.

Field says she has not found a consumer use that works end-to-end, and both companies are still being tested. She attributes that gap partly to incentive: businesses will retry a broken agent because it can save money or time, while consumers walk away the first time it fails. She cited a friend's anecdote that Muse had a bad day when capabilities got a lot worse, and for Dots she says the verdict is still out.