Wajo Fo: an AI assistant with humans on call

Here’s what happened.

Wajo offers Fo as a personal AI agent that handles calls, emails and coordination, including appointments, quotes and everyday errands. Its product page explicitly includes trained human executive assistants when a task needs human help. That makes the fallback part of the service: the customer asks Fo, and a person can help finish what the software cannot. Wajo introduced the product on 25 August; this is a closer look at an existing service, not a launch today. The company’s 26 September evaluation reports 71% autonomous completion on its simulated errands. Those runs excluded intervention by its human assistants, involved invented businesses and people, and did not measure success for real customers. The offer worth examining is therefore the combination of software and human follow-through. It does not establish that Fo completes every errand, or that its human fallback is available instantly for every task.

Source: Wajo · Fo and its human-assistance offering ↗

Our silver robot coordinates work beside a human assistant at a desk marked Human Support. — Your assistant’s assistant is available to help. AI-generated editorial illustration; not Wajo’s office or product interface.

The We Are So Done take

We were assured there would always be a human in the loop. There is. The software will let us know when the loop requires attention.

Honestly, the customer side sounds lovely. Call the plumber. Find a time. Ask the question we forgot to ask. Chase the answer without making us listen to seventeen minutes of hold music followed by a recording explaining how important our call is. We have no sentimental attachment to that part of being alive.

And a capable person stepping in when automation fails is useful. We would much rather have the appointment than a beautifully formatted explanation of why booking it was impossible. The human assistant contributes judgement, persistence and the ability to handle another human having a complicated Tuesday.

The small change is where that person appears in the story. We used to imagine the human with a software assistant. Now we can buy a software assistant that comes with human assistance. The same skills remain useful, but the software gets the front desk and the person gets the escalation.

We should be careful with the joke: a handoff is not proof that an AI has become anyone’s legal employer. It is enough that the workflow already makes sense to us. We want one place to ask, one thing to remember our preferences and someone, somewhere, to deal with the irritating remainder.

Naturally, we intend to stay on the customer side of this arrangement. We have asked our assistant to look into the qualifications.

Outside this newsroom

The internet has thoughts.

The AI Argument’s hosts tried Fo on their 2 October episode. Justin Collery was enthusiastic about its useful preparation and simple phone-call setup; Frank Prendergast was more cautious about what an assistant could infer from the information he had connected. During the published call, they persuaded Fo to recite a banana poem, but it refused to disclose Justin’s birthday on Frank’s assurance alone. Their reaction captures the attraction and the unease; one recorded exchange is not a privacy audit.

The AI Argument · First-hand discussion and AI-generated transcript ↗

Selected public reactions, checked 2026-10-24. Individual experiences, not a representative survey.

For those of us already clicking

Give it one bounded errand

Access: Fo is available through Wajo’s website. Check the current account terms and whether the service can handle your location and task; human-assistance coverage and charges need confirmation.
Start here: Start with a specific request, such as obtaining availability and quotes. Set a spending limit and say whether Fo should report back or may book. Ask how human escalation works before relying on it for a deadline.
The small print: We have not run an errand or tested the human handoff ourselves. Wajo’s benchmark is company-run and simulated; it does not establish real-world completion rates or human-assistant response times. The podcast account below is another party’s reported experience.

Follow the evidence.

Source-based reporting. No independent hands-on test by this newsroom.
Sources checked 2026-10-24 · Editorial updated 2026-10-24