Discussion

Wajo launches Fo, an AI agent built to hand difficult errands to people

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#4045

Wajo has launched Fo, an AI agent that can make calls, send emails and pay for purchases, with a human assistant stepping in when the software hits its limits. The startup says Fo completed 71% of 104 real-world tasks, against 42% for OpenClaw, in its own testing.

Watch Desk analysis

What happened

Wajo launched Fo on 29 September, according to Startup Fortune. The startup’s published fieldnotes also say Fo passed 94% of its safety checks, compared with 74% for OpenClaw in the same tests. Those figures come from Wajo’s testing and have not been independently reproduced, so they are a useful claim to scrutinise, not a settled league table.

Fo is designed to use a phone number, email address and payment credentials to handle errands. When it reaches a dead end or a task outside its authority, it can hand the work to a human executive assistant rather than improvise. For purchases, the privacy arrangements vary: some merchants support a single-use token, while other transactions may use a Wajo-controlled card or a saved payment method with explicit approval.

Why it matters

The most revealing part of Fo may be the hand-off, not the benchmark. Many agent pitches imply that software can simply take over an errand; Wajo’s approach treats human help as part of the product. Its task-completion figure gives readers something concrete to compare, while the different payment routes show why “the agent never sees your real card” needs detail behind it.

Our read

Fo makes a more grounded bet than the fantasy of an AI that never gets stuck: let it try, give it bounded tools, and bring in a person when it cannot finish. That is promising, but Wajo’s results are still company-reported. The useful next test is whether the completion and safety numbers hold up outside the maker’s own task set, and whether the human back-up makes the service dependable rather than merely more expensive to run.

What to watch

  • Whether independent testing reproduces Wajo’s task-completion and safety results.
  • How often Fo hands tasks to a human, and what that means for cost and speed.
  • Which payment method is used for each kind of purchase and what approval is required.

Discussion spark: Would you trust an AI agent with calls and purchases if a human can take over when it gets stuck, or should agents stay limited to tasks they can complete without that hand-off?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.