A Tool That Turns a Spoken Request Into a Work Handover
Voice requests can now connect to on-screen approvals and document work, opening a way to build industry-specific handover tools that prevent field instructions from being lost.
Published 2026. 9. 27.
Voice conversations are starting to move real work
ChatGPT Work, released on July 9, 2026, introduced capabilities that go beyond conversation to work with connected work tools and files. It can create or revise documents, spreadsheets, and presentations, and carry out needed tasks in a web browser.
On the web, iPhone, and Android, people can request this work by voice. If work remains after a call ends, they can view its progress and continue handling it in the text view of the same conversation.
The opening here comes not from a legal change, but from a change in product permissions and approval rules. Voice can receive a request, but its access to files or messages cannot exceed the existing permissions of the connected account or a company administrator's settings.
Actions that change an outcome, such as revising or sending a document, may require on-screen approval. A user saying they approve is not enough; they must directly select an approve or reject button.
The opportunity opened on July 9, 2026. Availability differs by account, subscription, and company settings. At launch, for example, the Work desktop flow did not include direct PowerPoint editing. That makes this a good moment to start with a narrow, industry-specific workflow rather than trying to replace every task.
How a day at a window-installation site could change
Consider a site manager at a window-installation company with 12 employees. In the morning, they visit apartment construction sites. In the afternoon, they check a measurement error and a shortage of materials at another site. While moving between locations, they take calls from workers and the office.
A worker sends a voice message saying that a handle is missing from unit 803 in building 101. A customer calls to ask for the silicone finishing to be checked again. The site manager pulls over, writes the details in a notebook, calls the office to schedule the work, and then copies the same information into a daily work log and settlement sheet that evening.
The same details get copied three or four times. If a building number, unit number, assignee, or deadline is missing, a worker may arrive the next day only to discover that a needed part is unavailable. It may also be unclear who heard the voice message.
When work can continue from voice, the manager can say while walking through the site: replace the handle in building 101, unit 803; assign it to technician Park; complete it by tomorrow morning. The service separates this into a task title, location, assignee, deadline, and required parts, then prepares to add it to the existing schedule and work log.
When the call ends, the text screen shows three new tasks, one task with missing information, and one schedule change awaiting approval. The manager corrects the wrongly recognized unit number, approves only the schedule change on screen, and hands the work over to the office staff.
Some work clearly remains with people. Site staff must judge whether a defect actually exists, negotiate schedules with customers, approve work that affects safety, and confirm the quality of completed work. Automatically generated work logs and cost sheets must also be checked before they are shared.
So the product to build is not another general-purpose voice assistant. It is a small handover screen that understands the different language and required checks of trades such as window installation, building equipment work, and in-home repair; asks again for missing details; and records who approved what.
Outside Korea, voice is being connected to work records
The ground-handling organization of Ireland-based Ryanair used Zello Work across 36 European airports. It set up more than 6,000 check-in, gate, baggage, and maintenance staff within a week, operating smartphone voice channels and message records instead of existing radios.
Zello uses a subscription model starting at about KRW 11,000 per person per month, or about KRW 9,300 per month with annual billing. According to a company case study, 300,000 voice messages were exchanged each month and incident investigation time fell by 70%. These are supplier-published customer results, so they do not guarantee the same outcome elsewhere. In 2026, Zello expanded into task-management features that create tasks by voice and let assignees accept them.
US real-estate software company MRI Software used Otter for customer meeting records and handovers between account owners. It stored meetings accumulated during sales cycles lasting one to two years, allowing a new owner to find summaries, follow-up tasks, and the context behind past decisions.
Otter publicly lists individual and team plans at roughly KRW 12,000 to KRW 29,000 per person per month, while plans for larger organizations require a separate quote. An Otter-published case study reported saving 20 minutes per meeting and recovering implementation costs in two and a half weeks, but those figures came from its MRI Software customer case study. The important point is that it did not sell voice accuracy alone. It also sold a record that lets the next person find the next action when responsibility changes.
What could be built from this
1. A field-instruction loss-prevention inbox
- What it does: Organizes calls and voice messages into a task name, location, assignee, and deadline, then asks again only for missing fields.
- Who uses it: A manager at a window-installation or building-equipment contractor who moves between three or more sites and relays instructions between technicians and the office.
- Why now: Voice requests can continue into connected schedules and document work, while actions that make changes can be held behind on-screen approval.
- First screen: Shows today's incoming voice requests in four groups: unreviewed, assignee needed, awaiting approval, and complete.
2. A shift-handover checker
- What it does: Extracts status changes and unfinished work that the next shift must know from a spoken end-of-shift report, then drafts a handover note.
- Who uses it: A manager at a small care facility where day and night staff must hand over resident status and facility-check information.
- Why now: Spoken information can feed into document drafts and follow-up checks, though access permissions and review procedures for sensitive information must be designed separately.
- First screen: Puts changes by handover subject, items not yet reviewed, and confirmation by the next shift first.
3. A visit-repair completion report maker
- What it does: Splits a technician's spoken post-job account of the fault cause, replaced parts, photos, and need for a return visit into a customer report and an internal settlement record.
- Who uses it: A boiler or water-purifier repair technician who visits several homes a day and writes reports in a batch at night.
- Why now: A technician can create a voice draft on site, then check photos and amounts in the text screen after the call.
- First screen: Shows each visit scheduled for today alongside report complete, photos missing, cost check needed, and awaiting customer delivery.
4. An approval helper for workers who find hand use difficult
- What it does: Takes voice requests for document drafts and information organization, then collects only the steps requiring a real change into large approval buttons.
- Who uses it: Office workers who have mobility or hand-use limitations that make prolonged mouse and keyboard use difficult, along with the people who support them.
- Why now: Work can continue after a voice request while final approval remains visible on screen.
- First screen: Arranges in-progress work, items needing content review, approvals needed, and failed work with large text and simple actions.
What to check today
Call one person who does field work and ask them to describe just five voice instructions they received during a recent day, in time order. If, within 30 minutes, they identify even one case where the same information was copied into two or more places or where they had to contact someone again because an assignee, deadline, or location was missing, that is a strong first problem to build for.
Why this matters where you are
Check whether field teams in your market repeat the same spoken instructions across calls, messages, schedules, and reports. The tools, approval rules, and privacy requirements may differ, but the handover gap is visible in the missing details, repeated data entry, and unclear ownership of the next action. Start by mapping one narrow workflow and the fields that cannot be left blank.
Sources
7 sources
Every fact in this article came from the pages below. Check them yourself.
- Official announcement on using ChatGPT for complex workOpenAIUsed for the release scope on July 9, 2026, and examples of connectable work tools.https://openai.com/index/chatgpt-for-your-most-ambitious-work/
- ChatGPT Work and Codex Help CenterOpenAIUsed to confirm voice environments, continued task handling, and the limits of on-screen approval.https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex
- ChatGPT Voice Help CenterOpenAIUsed for the scope of voice conversations and connected-app use.https://help.openai.com/en/articles/20001274-chatgpt-voice
- ChatGPT connected apps Help CenterOpenAIUsed for the explanation that existing account permissions and company settings continue to apply.https://help.openai.com/en/articles/11487775-connected-apps-in-chatgpt
- Creating and editing documents, spreadsheets, and presentations in WorkOpenAIUsed for supported outputs, PowerPoint limitations, and the areas that require human review.https://help.openai.com/en/articles/20001278-creating-and-editing-documents-spreadsheets-and-presentations-with-chatgpt-work
- MRI Software customer story with OtterOtterUsed for the handover approach for long sales records and time saved per meeting. This is a supplier-published customer case study.https://otter.ai/case-study/mri-software
- Otter pricingOtterUsed for public subscription prices for individual and team plans and the separate-quote model for enterprise plans.https://otter.ai/pricing?msockid=089dc62b127f6de8098fd06113c76c41