What is a natural language interface?
A natural language interface is a way to use software by typing or saying what you want in everyday words, instead of finding the right menu, form or command. The system works out what you mean and then answers or carries out the request.
How is a natural language interface different from menus and forms?
A menu-and-form interface asks the user to translate their goal into the software's structure. They need to know which screen holds the setting, which filter narrows the list and which button commits the change. A natural language interface reverses that: the user states the goal, and the software does the translating.
Take an expense tool. A manager who wants to approve every expense under $50 from their team this week has to open Approvals, filter by team, set a date range, sort by amount, select the rows and click Approve. With a natural language interface, they type "approve my team's expenses under $50 from this week" and review what comes back.
The difference shows up most in three places:
- Discovery. Menus only help if you know the feature exists and what it's called. Plain words work even when the user doesn't know the product's vocabulary.
- Multi-step tasks. A request can span several screens at once, where a form handles one screen at a time.
- Infrequent tasks. Nobody remembers where the annual export lives. Saying "export last year's invoices as CSV" needs no memory of the path.
Menus still win where users scan, compare or make precise visual edits. Most products keep both and let the user pick.
Familiar examples include voice assistants, search boxes that accept full questions, tools that turn a plain question into a database query, and agents inside business software.
How does a natural language interface work?
Older natural language interfaces relied on keywords, fixed phrases or hand-written grammars, which broke as soon as someone phrased a request differently. Modern ones use a large language model to interpret the request, usually in four steps:
- Interpret. Work out the intent and the details: which records, which dates, which amount.
- Ground. Check the request against what the system knows right now, such as who the user is, what's on screen and what data exists.
- Respond or act. Return an answer, or call the product's own functions to carry out the change.
- Confirm. Show the result, and ask before anything that's hard to reverse.
Step 3 is where natural language interfaces split. Some only answer questions about the product. Others take the action. A search box that understands full sentences is a natural language interface, and so is an agent that books the meeting you asked for, but they solve very different problems.
Where does a natural language interface fit in a product?
It fits wherever the gap between what users want and where the controls live is widest: dense admin settings, reporting, bulk edits, first-time setup, and anything a user does once a quarter.
The hard part isn't understanding the sentence. It's acting on it safely. That means the interface has to run through the same rules as the rest of the product: the signed-in user's permissions, the product's validation and its audit trail. A request the user isn't allowed to do by hand shouldn't succeed by typing it.
This is the approach Foldspace takes: a typed request gets no extra power, because the Product Agent carries it out in the product with the signed-in user's own permissions, through the backend APIs the product already has.