Accessible Intelligence

When you ask a colleague about a building, you rarely explain which folder to open or which filter to apply. In an application, those steps often become your responsibility: you know what you need, yet still have to learn how to ask through menus and buttons.

Knowing where to click has become a professional skill. Someone in the office knows how to get a particular report out of the system; everyone else asks that person.

In this blog post, I want to explore how applications can adapt to us, from the way we ask to the tools we use.

A colleague who always answers with a spreadsheet

Imagine an analyst who answers every request with the same spreadsheet. You ask which buildings need attention; she sends the full table. You ask for a short explanation; she sends it again. The information is correct, but making it useful remains your responsibility.

In practice, a colleague adjusts: a detailed comparison for an analyst, a recommendation for the committee, an explanation of an unfamiliar figure. What she knows stays the same; how she helps depends on what you need.

An application's interface (its screens, controls and conversation) determines how we use it. Designers traditionally prepare these options in advance; an adaptable interface lets the arrangement follow the person and the task.

Talking to the computer is already possible

With Codex's voice mode, you can already speak, hear replies and discuss an assignment while it runs: check progress, clarify a request or change direction. As in asynchronous intelligence, work continues and you return to the conversation when needed.

The result should also suit the task. Nobody wants fifty building names read aloud when a map would make the comparison clear.

An interface you can shape

A generative interface (often called gen-UI) is a screen or tool assembled around what you ask for. An agent selects and connects elements such as tables, maps, filters and timelines, using the available information. You can describe how you want to work, including a view its designers never prepared as an option.

That might mean a compact table with every source close at hand, or a map with a few explanations and only the decisions requiring attention. You shape the view through requests and refinements, then decide which recurring choices the application should remember. The layout, level of detail and controls can follow each person's needs and preferences.

Léa asks for her own working page

Return to Léa, our portfolio analyst. Her team divides and checks the work, continues while she is elsewhere and builds information it can reuse. It has now compared the renovation works that could affect next year's rental income.

She opens the review and asks:

“Show me the buildings where approved works overlap with a lease expiry. Keep the unresolved cases visible.”

The agent prepares a page with buildings grouped by region, a timeline and the rental income concerned. Léa refines it: “Put the timeline first, keep the sources beside each figure, and let me filter by renovation stage.” She keeps this version for her morning review, with checked updates as new information arrives.

For the committee, she could request a tool comparing two renovation schedules, with adjustable assumptions and their consequences side by side. Agents assembling views or writing their code reduce development costs enough to make a tool worth creating for a single decision.

Accessibility for each person

Her manager keeps a list of decisions and who owns them. Another analyst prefers a map, even though she has the same role as Léa. A committee member reads in French; a new colleague asks for explanations of unfamiliar terms. They use the same checked information through views they can shape independently.

Someone with low vision may prefer an enlarged table or a spoken explanation. A colleague who cannot comfortably use a mouse needs to complete the review with a keyboard or voice commands. Each should retain access to the sources and decisions their role requires.

Léa's own needs change throughout the day: a spoken update on her phone, a comparison at her desk, a shared committee view. The assignment should follow across them. These interfaces still require testing with assistive tools and their users; generating a page does not guarantee accessibility.

What must not move

An application that keeps moving familiar controls becomes harder to use. It might offer Léa her usual lease timeline as a starting view, but she should be free to decline or undo the change.

The facts need the same care. A budget awaiting approval keeps that status in a table, a spoken answer or a summary. Disputed figures remain marked, access restrictions still apply, and discussing an expenditure never counts as authorising it.

The organisation behind these views makes that possible: shared definitions, reviewed calculations and clear responsibilities keep the work consistent as its presentation changes.

Making intelligence accessible

Across these four articles, intelligence has gained an organisation, time to work and information to build on. AI-powered applications give people access to those capabilities through conversation, voice, a working page or a tool assembled for a particular decision. These forms should keep widening as people's needs do.

What matters to me is the leverage this gives someone like Léa: how much she can investigate, understand and carry forward with the organisation working alongside her. A well-designed dialogue box is one way to reach it; the ambition extends to everything the application helps her accomplish, whichever way she chooses to interact with it.