The interface belongs in the workflow

Project Catalyst | Part 2 of 4

Routing everything through the AI chat is one way into capital markets workflows. Keeping the screen is another. But the two aren't mutually exclusive.

Software vendors in capital markets are under growing pressure to open up their data and analysis through MCP (Model Context Protocol, the emerging standard for letting AI reach into software).

The workflow is moving into AI, and a product the AI can't see might get dropped altogether.

Most vendors expose some of their data and analysis via API, but many chose long ago not to be API-first. They spent the money on an interface instead, because the interface is what the customer buys and what the customer trusts.

The pressure now is to reverse that decision, but fears of the SaaSpocalypse have many vendors forced to choose between two extremes for how to ‘adopt AI’.

The first is to build the AI into their own product, a chat bar in the corner of their own screen running on their data. One vendor described the ceiling on his own AI work without being asked: it answers anything inside his data universe and nothing outside it. No email, no research from elsewhere, no order sitting in another system.

The second is to open everything through MCP and let the customer's AI do the rest. The product becomes a supplier to a chat window it doesn't own. You're as relevant as the prompts people happen to write, and the customer can't tell how much of any answer came from you. The desktop real estate and the mindshare that came with it are gone.

Some clients still prefer to point and click. Not every workflow suits prompting and reading words. In capital markets the complexity on a screen doesn't translate to a wall of text, and when the margin for error is this thin, people often want a literal finger on a literal button.

AI can generate visuals on the fly, but if the outcome varies each time, it defeats the muscle memory a trader spends years building. When the rendering is wrong, the person looking at it has no way to know, because the vendor is the one who understands the shape of the data.
So the standoff holds. Clients want AI that makes their people more productive. Vendors own the analysis and won't hand over the surface it lives on.

But there's a third option that keeps the screen and joins the workflow. The AI opens the vendor's own interface with the answer already on it, next to whatever else the question needs.

It's the vendor's UI, inside the customer's AI workflow.

The AI opens the screen with the source of truth

A desk asks its AI a question. The AI works out which applications can answer it, opens the vendor's screen with the answer already there, and arranges whatever else the question needs around it. We call that invocation, and what it delivers is intelligence at the point of action. The AI sees what the user sees, and can take action back into the apps to the point where human decisions are required.

The case for this method is laid out in our first post: The right screen beats the right sentence. The answers a desk needs are already on screens it trusts, and the AI's job is to open them and work through them with the user.

The work on the vendor's side is small. The license stays, so does the customer relationship and the commercial terms. A vendor still makes decisions application by application and customer by customer.

An MCP strategy still makes sense when invocation runs beside it.

Nothing about the way vendors sell today has to change

Distribution here is a channel rather than a replacement. A vendor sells the way it sells now.

What the channel adds is presence. Desktop real estate is mindshare, and a screen sitting in front of a trader all day is where a vendor's brand lives and where its next idea gets noticed.

What keeps a good tool underused is the cost of switching windows, and new users never learning what the desk's best people already worked out. A tool or intelligence that arrives when the work calls for it beats one that waits to be remembered.

Without the screen, a vendor's contribution can sit buried far enough inside the final answer that somebody at a renewal meeting might ask why the firm is paying for a thing they have never seen.

Buyers search by the problem they have

Distribution is driven by the catalogue. Inside a firm's own environment, applications get profiles based on how real users trigger them. That knowledge stays in the firm's workflow memory, and it drives future searches through the store to the apps best suited to those pain points.

For a vendor that means getting found for a capability it might not have otherwise led with.

The middle has to be neutral

If the environment in question is owned by a competitor, the vendor has swapped handing over its data to handing over its position. That is a worse trade than the one it was avoiding.

HERE, formerly OpenFin, has held the neutral position in the capital markets desktop for more than fifteen years. Project Catalyst takes that seat deliberately. The third option only exists if the place where it happens isn't itself a walled garden either.

One integration instead of thirty

Application developers also gain by what they lose. One to one integrations and bespoke client requests take time and effort better spent on broad product development.

A desk we sat with had built its own signal for how hard a name would be to trade. They owned the intelligence and paid for the system it needed to reach. The integration took six months, because the same vendor had thirty other clients asking for something slightly different and could build one of them.

On a shared environment those thirty requests become one, built once against something any firm on it can reach.

The decision is the one already made

A screen carries what text can't, and people trust it. Neither of those has changed. What's new is that a firm's trusted AI system gets the tools to decide which screen gets opened.

The user and the AI end up on the same screen, working from the same context, and the person doing the job keeps their hands on it.

The capital markets desktop is being rebuilt around AI right now, and the best version of it gets built together with the vendors whose applications traders already rely on.

Get involved

We're running pilots now, inside firms' own environments, on their own models. If you're building something a desk should be able to reach, talk to us.

Details are live at here.io/project-catalyst. Email us at catalyst@here.io.