From a conversation to a completed assignment
– · 2 min read
A retrospective selection, researched on .
An AI can give an answer. Or it can turn several sources into work you can inspect. This week, OpenAI changes both sides: the conversation and the connected assignment.
Two changes in two days
On 8 July, OpenAI announces GPT-Live-1 and its mini variant for ChatGPT Voice, able to listen and speak simultaneously. On 9 July, ChatGPT Work follows for longer assignments and finished artifacts. Access and rollout depend on the plan and workspace. Dated release notes.
An assignment that makes the difference visible
Our example: “Compare these three quotations and prepare a decision brief.” That assignment receives the quotations, selection criteria and permitted sources. It needs to extract prices and services, mark differences and produce a readable brief. The output would be a table with a reasoned comparison. A beautiful table alone does not show whether a hidden additional charge was correctly read.
The work consists of several connected steps. You can answer questions, redirect the assignment and separately approve important actions. That ongoing collaboration belongs to the announced Work format; the quotation comparison is our own example.
The new question is: what exists at the end?
A more fluid conversation can make coordination easier. A longer assignment also needs a clear endpoint: which file, decision brief or checkable change do we expect? This distinction helps in practice. Define the input and output, then inspect a critical detail — for quotations, the same service and its full cost. That turns “the answer sounds good” into a reviewable acceptance check.
Assess an assignment by its expected output and critical details. Conversation quality and successful work need different checks.