The GUI is Dead.
Long Live the GUI.
by Jacob Koenig
3/3/26
The SaaSpocalypse will kill the point-and-click menu, but the GUI is a different story.
Will AI Agents Actually Replace Software Interfaces?
There’s a thesis gaining traction in tech circles right now that says AI agents will eat all of SaaS. The point-and-click interface is dead, so every piece of vertical software becomes a headless API, and agents handle the rest.
The direction is clearly right. The era of navigating six nested menus to find one setting is ending, and we’re all better for it. AI agents will absorb the clunky, multi-step workflows that make most enterprise software painful to learn and taxing to use.
But the idea that GUIs go away entirely? That visual interfaces become irrelevant because an agent can just tell you the answer? That’s a builder’s fantasy, not a user’s reality.
Switching Costs Are Human, Not Technical
The blind spot in the “no more GUIs” argument: it confuses what the agent does with what the human experiences. An AI agent doesn’t feel switching costs and it doesn’t care whether it’s querying Bloomberg or a competing API. To the agent, every system is just an endpoint. But the end user is still interacting with a UX, and users of mission-critical software won’t abandon their muscle memory because someone shipped a better chatbot.
A portfolio manager who’s lived in Bloomberg for 15 years has fingers that know the keyboard shortcuts and eyes trained on where the numbers that matter show up. That accumulated instinct is what lets them make million-dollar decisions under time pressure. That’s professional fluency, and it took years to build.
Some technical founders already live in chat interfaces and voice commands for everything, and they look at a traditional dashboard and see a relic. But most people in most industries aren’t there, and they won’t be for years.
Most professionals still want to see what they’re used to seeing, where they expect to see it. They want familiarity with intelligence layered on top.
Why Don’t Most People Build Their Own Interfaces?
As someone who has dedicated months vibe-coding my own to-do list app, I know the cost of getting the exact UX you want. I had specific visualizations, specific workflows, specific views of my own data in my mind, and I had drive and enough technical fluency to go build it.
Most people don’t have an idea of what they want until they see it presented to them. They think in screens they recognize. Asking them to start from a blank AI canvas and describe what they need to see is like handing someone an empty plot of land when they ask for a house.
The better move is preserving the components people already know while imbuing those components with dynamic intelligence.
Don’t Replace the Screens. Orchestrate Them.
The future lives somewhere between a million disconnected GUIs and zero GUIs: familiar visual components pulled from systems people already use, assembled dynamically by an AI layer that understands what you need to see and when you need to see it.
The unlock is a system that can pull in the components you’re familiar with into a centralized window, overlaid with an AI interface. That’s the zero-switching-cost argument. You’re not forcing anyone to learn a new paradigm, you’re giving them what they already know how to read. It’s stripping down the silos that made them code-switch between five applications just to get a complete picture.
A chart showing portfolio exposure over time communicates in a glance what would take paragraphs to describe in text. But “why did our sector allocation drift this week?” is a question, and the AI can answer it with context your static dashboard never could. Visual interfaces and conversational AI serve different cognitive needs, and the system should know which to use when. That’s the GUI getting promoted from a display into a collaboration surface.
Where This Matters Most: Execution Software and Muscle Memory
Capital markets present the best example why visual interfaces aren’t going anywhere. A trader watching their execution management system is tracking live orders, reading market depth, and watching fills populate in real time. Timing can mean the difference between a good fill and a million-dollar miss, and that kind of situational awareness is fundamentally visual, if not multi-modal. Muscle memory here translates to alpha, and the human switching costs come with real risk.
But you can certainly make what they’re already watching smarter. The screens stay and so does the blotter and the fills view. The difference is an intelligence layer on top that triages what deserves attention, flags anomalies, and can explain why when you ask. The progression goes from passive oversight to active coordination.
The institutional demand signal already exists. Every buy-side desk is asking their vendors “what are you doing with AI?” and the vendors are all “looking to provide” solutions. They’re retrofitting legacy systems with AI as a feature rather than rethinking the architecture. The actual opportunity is consolidating the existing tech stack and adding intelligence across it.
The Real Question Isn’t “Will GUIs Survive?”
The real question is whether the software industry will evolve with its users or try to drag them into a paradigm they didn’t ask for. People accept intelligence that makes their existing workflow better long before they’ll trust a fundamentally new one.
The headless pipes need a brain, and the visual layer needs awareness. But awareness isn’t automation, and intelligence delivered in context isn’t the same as intelligence requested through a prompt. The best version of this future meets people exactly where they are and makes what they already see dramatically more useful.
Nicolas Bustamante of FinTool recently laid out a framework describing which vertical software moats survive the AI era and which ones dissolve. His three-question test comes down to proprietary data, regulatory lock-in, and transaction embedding. If the software holds data that doesn’t exist anywhere else, satisfies compliance requirements that can’t be ported cleanly, or sits directly in the path where money moves, an AI agent can’t simply route around it. Those are exactly the moats held by the entrenched SaaS players that capital markets professionals already depend on.
Rishi Nangalia of OPCO makes a complementary point: enterprise consumers can finally take power into their own hands to extract value across vendors and applications without building large technology teams or deep in-house expertise. The tools that serve them best won't replace the vendor stack but sit on top and consolidate what's already there.
In Institutional Fintech, the solution isn’t to replace the screens. Instead, componentize the existing software stack, work with those incumbents, and imbue intelligence across all of it.
Jacob Koenig builds AI intelligence systems for capital markets. Previously Goldman Sachs (12 years, Head of Taiwan Execution Services) and Partner at Woodbridge International.
jkoenig@komcp.com
This article was also posted separately on LinkedIn: https://www.linkedin.com/pulse/gui-dead-long-live-jacob-koenig-slsse/