Issue 002 · 6 min read
Three separate changes in one week each removed a barrier to doing something: verifying a voice, shipping an application, attempting another profession's task. None of them moved the responsibility for the result, which stayed exactly where it was.
Skip if you like
An AI assistant is a tool you talk to in plain language, like a fast, very literal colleague who has read a great deal but knows nothing about your job until you tell it. It can draft, summarise and suggest, but it does not decide, and it can be confidently wrong.
A prompt is simply what you type or say to the assistant: your request, plus any background and examples you give it. The clearer and more specific the prompt, the more useful the answer, so treat it like briefing a new starter, not typing a search.
A context window is how much the assistant can hold in mind at once: your instructions, the document you pasted, and the conversation so far. When a chat gets long it starts to forget the earliest parts, so for a big task give it the key facts again rather than assume it remembers.
Start here
Newly announced
This establishes where a file came from. It does not establish that the words are true, that the speaker agreed, or that you may publish it — and a missing signal is not proof a file is genuine.
See the detail and the sources →Newly announced
Construction was never the hard part of an internal tool; deciding what correct means was. When a prototype costs an afternoon, the specification becomes the whole job.
See the detail and the sources →Newly announced
Access to a task is not competence at it, and neither is authority to approve it. The study describes what people asked for, not whether the answers were any good.
See the detail and the sources →
The one that mattersWhat's genuinely new
OpenAI measured the crossover rather than asserting it: more than 800,000 work-related messages across eight occupation groups, with the occupation-specific share reported separately from the generic one.
What the makers claim
The framing around all three is capability — what the tool can now do. Provider material does not describe output quality, review coverage, or who was accountable when the output was wrong, and the crossover study measures message patterns rather than results.
Where it helps
Where you already own the outcome and the tool removes the grind: you know what a good answer looks like and were simply slow at producing it.
Where it doesn't
Where you could not tell that the output was wrong. If you could not have caught the mistake before AI, you cannot catch it now — you can only produce more of it, faster.
What remains yours
OpenAI classified 16.8% of all work-related messages, and 43.5% of non-generic occupation-specific messages, as tasks historically associated with another occupation.
Take one AI-assisted task that has crossed into a new role and name, in writing, the accountable owner, the specialist reviewer and the trigger that escalates.
Try this
A task that used to go to someone else is now something you do yourself with AI, and nobody has said who checks it.
Prepare the first version, list the assumptions it made, and mark the parts it could not verify from the material you gave it.
Decide what wrong would look like in this task, and whether you would recognise it.
Name one AI-assisted task you have taken on in the last month that used to belong to another role
Write one sentence describing what specifically wrong looks like in it, not merely bad
Answer honestly whether you would notice that failure if it happened
If the answer is no, name the person who would, and route the next one past them before it goes out
Step three only works if answered honestly. The purpose is not to stop doing the task, but to know which parts of it you are guessing at.
The Human Lens · what stays yours
For most of working life those two questions arrived together: if you could do something, you had usually earned the right to. AI has separated them. The tools answer the first brilliantly and have nothing to say about the second, which makes asking it out loud a larger part of the job than it used to be.
Worth watching
One question
See you next Monday.
Bring it back to your work
Bit starts from this issue, then asks you enough about your own work to make it worth acting on. If the honest answer is that none of this touches your job yet, it will tell you that too.