We wanted voice AI in our payments app. Here’s what we cut.
Aakash Mahadevan · 31 October 2026
Read the originalGlomo's product design team cut a planned voice payment assistant down to one job, repeating a past payment, and kept the final approval on screen with a face check.
Aakash Mahadevan, a product designer at Glomo, a cross-border payments company, describes a voice AI assistant planned for the Glomo account, an app for holding a balance and paying people. A hard design review before any code was written found that spoken payment requests were ambiguous and that a screen-free design was unsafe. The team rebuilt the idea around one rule: speech prepares a payment and the screen confirms it.
- One example request, "Pay Rahul $1,000 using HDFC Bank", could point to several people named Rahul and to either of two bank accounts.
- As the assistant listens, a payment card fills in on screen, and the payee, exact amount, fee and Face ID approval always happen on screen.
- The first version only repeats payments to saved payees from the Glomo balance, with a cap per payment, identity checks before the microphone opens and English only.
- Before building, the team plans a Wizard of Oz test with eight to ten customers, with a target of zero payments to the wrong person.
A design team adding an AI assistant to a risky action can let the assistant prepare the action and keep the approval on a screen the person reads. The design has not yet been tested with customers.
Write down what the first version of an AI feature will not do before deciding what it will do.
Run a hard design review of an AI feature before any code is written, testing each part of the concept against what happens when it goes wrong.
Derived by Working Surface from the article; more in the Playbook. Source line: Your voice never approves a payment.
1 November 2026: Teams putting AI agents to work decide in advance which steps a person must still take, and that person carries more review and a greater duty to push back.