Asking well
The assistant builds what you describe. A few habits get you the right application the first time, and a shorter conversation.
On this page
Use your own words
No technical vocabulary is needed: describe your work as you would to a new colleague. Say what the application is for, who will use it and what it must manage. For example:
We are a 60-person construction company. I want to track our subcontractors' certificates: the social security compliance certificate, the company registration extract and the ten-year liability insurance certificate. For each subcontractor, we record the name, the company number, a contact, and for each document its expiry date. The buyers (3 people) must see at a glance what expires within 30 days.
Started from a template, the application already does the essentials: describe what is different for you, not what it already does.
Say who does what
Say who enters data, who only looks, and who creates colleagues' accounts. Otherwise the assistant will ask: answering in your first message shortens the exchange.
Start from a real case
The spreadsheet you use today, a sample document or a screenshot of an existing tool say more than a description. Attach them with the paperclip, “Attach files”: images, PDFs, spreadsheets, documents or ZIP archives, several at a time. You can also drop them on the message box, or paste a screenshot into it. A pasted text of 4,000 characters or more, such as a long specification, becomes an attached file too, “pasted-1.txt”, which the assistant reads in full. A correction takes none: send it in a message.
A message you started and have not sent is not lost: come back to the conversation, even after closing the browser, and it is in the message box again, with its attached files and its “Point” labels. It stays on this device, in this browser, until you send it, empty the box or click “Sign out”.
An attached image can also go into the application: your logo, an SVG one too, a photo of your team or of a product. Say so in your message. For an illustration, just describe it (“a beach at sunset”): the assistant picks it from a free image bank.
Only include the data that is needed: an extract is enough, and made-up names do just as well. The notice above the message box says where your messages are processed.
Another service: its keys stay yours
If the application must use another service, such as online payment, a map or text messages, the plan says so and names what that service will give you: a key, an account ID. The assistant never writes it into the application. It adds a field to its settings, the “Settings” page its administrators alone can open, most often in the “Configuration” section, and tells you which one to fill in. You paste the key there once the change is published, and change it whenever you like, without a new request. A secret key is no longer shown once saved: to change it, paste the new one. While the field is empty, the feature says so instead of stopping.
Answer the questions, then reread the summary
The assistant asks one question at a time, often with numbered choices: “The 2nd” is a full answer. Before each question, it sums up what it understood. Reread that summary: if a point is wrong, correct it right away, in your next answer.
Request the plan once everything is said
Once the assistant has understood, click “Request the plan”. A last detail written in the message box goes along with the request: no need to send it first. An attached file, however, is sent with “Send” before you request the plan.
Reread the plan before executing it
The plan describes in plain words what the application will do, what is left for later, then “How to check”: the tests to run once the application is built. When the assistant made a choice of its own that you had not stated, a line “Decided without asking you” says so at the end of what will change: read it first. Writing it takes a few minutes; you can close the page meanwhile.
To change the plan, write what should change, then click “Request the plan” again: only the latest plan is executed. Nothing is built until you click “Execute the plan”.
If the application already does what you ask, the assistant proposes no plan: it tells you so in a sentence or two, with what already does it. There is then nothing to execute, and on the Kanban the card does not move to “Plan ready”.
When a plan is waiting to be read, the Applications page says “Plan ready to read” on the line of the app or of its test version; a click opens the conversation right on the plan.
A touch-up afterwards
To adjust what was just done (a text, a label, the order of a list), write it, then click “Adjust directly”: the assistant applies it at once, without a new plan. If you have just discussed it with the assistant, no need to write it again: click the button with the box empty, and what you asked since is applied. The tests are then run again before saving, which takes a few minutes. For a larger change, send it and request a new plan.
While the test version is being updated after a change, the button stays clickable: write the adjustment, and it waits above the box and goes ahead by itself as soon as possible. “Cancel” takes it back. While it waits, it must be written out: an empty box, attached files and pointed elements wait for the update to finish.
One change per conversation
Once a change is done, start the next one with “New”, at the head of the conversation: the assistant starts afresh from the updated application, faster, and counting less towards the assistant’s usage.