AI Leadership Weekly · Issue #107 · Tuesday 27 October 2026 · 08:00 GMT

Hello.

A caller asks a service desk to move a maintenance visit from Tuesday to Thursday, then corrects the site address. What would convince you the call went well? The recording might sound flawless while the engineer still turns up on Tuesday. Three developments this week make that gap between conversation and delivery worth investigating.

IN 60 SECONDS

  • Mistral’s Voxtral 3 combines speech recognition, speech generation and tools for building voice agents. Test what reaches the booking record after a caller changes their mind.

  • Innovate UK opened a competition for operational AI. A supplier needs an eligible UK deployment partner and a real setting in which to demonstrate the work.

  • The Agentic AI Foundation took on four open projects. Ask which part of your team’s process you could actually reuse if you changed suppliers.

This week’s decision. Pick one service request that ends in a verifiable record, such as a rescheduled visit. Agree with the person delivering the service what a successful handover looks like.

TOP STORIES

1. Voice AI gets a harder test at the service desk

Mistral released Voxtral 3 speech models alongside a developer kit for building voice agents that interact with business systems while a conversation is in progress. Mistral offers an EU-hosted service and a deployment option for organisations using their own infrastructure. Buyers can ask their supplier to show where audio and transcripts go in the proposed setup and who can retrieve them. Mistral’s Voxtral 3 release · 20 October

Bring a dispatcher to the demonstration. Let them dictate a booking reference, interrupt to correct a digit, and ask for the final time to be read back. Give a second colleague only the resulting record. Can they send the right engineer to the right address on the right day?

2. A UK funding call puts the deployer in the room

Innovate UK opened the Operational Systems strand of its BridgeAI Deployment Catalyst. It offers a competitive share of up to £5 million for phase one. The lead must be an eligible UK-registered SME with a working AI or machine-learning product. The consortium must also include a qualifying deployer in an eligible sector. The programme asks for a demonstration using representative data and systems. Innovate UK competition scope and eligibility · opened 21 October

For a potential deployment partner, the first budget question is whose time the demonstration needs. Ask the service manager to cost preparing example bookings, making a test system available, supervising staff and checking completed jobs. Those hours belong in the proposal alongside the supplier’s engineering estimate.

3. Open projects give buyers something specific to test

The Agentic AI Foundation announced four additions, including Agent Skills for packaging instructions and resources, and Open Responses for describing interactions with models from different providers. The other two projects, OpenEnv and OpenSandbox, address training environments and execution environments respectively. Linux Foundation announcement · 22 October

Take your existing appointment-change procedure to two suppliers. Ask each to identify which instructions they can import, which system connections need rebuilding and which behaviours need retesting. The difference will help procurement estimate the real work of switching later.

CEO / COO / CXO CHECKLIST

  • CEO — Agree how many correctly fulfilled visits would justify the trial cost.

  • COO — Name the person checking that the old slot was released and the new one reserved.

  • CXO — Listen to how the caller gets a person when a correction remains unresolved.

THE 15-MINUTE PLAYBOOK

Build a three-call acceptance test. In the first five minutes, write three fictional requests on cards. Use a straightforward change, a corrected address and a caller who hangs up before confirming. In the next five, have the dispatcher write the expected booking record for each, including what should stay untouched. Use the last five to compare those records with the acceptance criteria in your supplier brief. Add any missing field and identify who will inspect it. The cards can then serve as the opening exercise at the next demonstration.

QUESTION FOR READERS

Which single field would your dispatcher check first before sending someone out on a job booked by AI?

Brought to you by Data Wave — your AI & Data Team as a Subscription.