Bamboo Connect Chief Strategy and Technology Officer, Joy Sengupta, explores how the rise of AI agents is challenging decades of software design thinking. We explore why the future of enterprise technology may be built around conversation rather than screens.
During a recent redevelopment at Bamboo Connect, I found myself back in familiar workshops shaping UI and UX. This wasn’t unfamiliar territory; in fact, I used to genuinely enjoy these kinds of discussions. The team was hands-on and creative, always focused on shaping the exact way a user experienced our system.
But this time, something felt fundamentally off.
The “60-Screen” Wake-Up Call
The moment of pure frustration came when we were mapping out what should have been a simple Quality Assurance (QA) process for our insurance-claims digital devices. This is a core part of what we do: a technician checks a device, logs its condition, and triggers the next step in the workflow.
Somehow, that straightforward sequence translated into a 60-screen UI plan. Sixty screens for a single operational task.
Despite everyone’s best efforts, the design conversations felt slow and unnecessarily complex. We were spending hours figuring out how to map out interactions that, in 2026, should really just happen in the background. Tasks our staff know instinctively still had to be painstakingly translated into wireframes, mock-ups, and rigid button paths.
It made me wonder: why are we still designing complex systems that users have to learn to navigate, instead of building systems that can simply understand us?
A New Kind of Interface – The Post-UX World
The tech industry has spent years championing “headless” architectures—decoupling the front end from the back end so the two can evolve independently. It was a sensible step forward, but autonomous agents take that logic even further – they skip the front end altogether. You don’t need a screen to tell a system what you want if you can just… tell it. If headless was the halfway house, what comes next is UI-less.
This shift moves us past basic chatbots and into the realm of truly autonomous enterprise agents. Sholto Douglas, a DeepMind alum now at Anthropic, recently described the experience of working with agentic tools like Claude Code as “utterly wild”. In this “Post-UX” world, our goal shouldn’t be to build a prettier dashboard; it should be to eliminate the need for one entirely.
If software can finally reason, forcing users to click through dozens of screens to provide data the system should already be able to deduce feels completely outdated. With modern tools capable of chaining reasoning, retrieving structured data, and triggering system actions, AI agents can execute complex workflows rather than just answer static questions. They don’t just sit there waiting politely for a prompt. They can hold a codebase, an inventory, and a stack of legal contracts in mind at the same time – and just get on with it.
The “Swarm of Bees” Architecture
Instead of one massive, clunky software platform, enterprise architecture is shifting toward something resembling swarm intelligence. Think of it like a beehive. Where legacy systems function like centralized, rigid mainframes, these AI agents work like a decentralized swarm: specialized, autonomous, and constantly in sync.
Instead of a single “brain” trying to manage everything, you deploy specialized AI agents working in parallel:
The Scout Agent: Scans incoming claims for data anomalies.
The Compliance Agent: Ensures every repair meets regional regulatory requirements.
The Logistics Agent: Coordinates replacement parts and shipping schedules automatically.
Like a swarm of bees, these agents coordinate via background protocols, not pixels. This background orchestration completely replaces the 60-screen nightmare. The technician no longer has to navigate a digital maze; they simply use the most natural interface we have. As NVIDIA CEO Jensen Huang famously put it, “The new programming language is human”.
And this isn’t niche. Gartner predicts up to 40% of enterprise applications will feature task-specific agents by the end of 2026, up from less than 5% in 2025. Dharmesh Shah, CTO of HubSpot, put it simply: “Agents are the new apps”. More than 80% of forward-looking enterprises now report that agents are actively replacing traditional packaged software as their main system of work. Across claims, logistics, and supply chains – anywhere the business logic is well understood but the day-to-day process is still painfully manual – this is already happening.
What This Might Look Like
Let’s look at a real-world scenario. Instead of a Bamboo Connect claims handler logging into three separate dashboards and making complex manual judgment calls based on courier histories, stock levels, and SLA constraints, they might soon just say:
“Do we have a device to replace a Samsung S22 for a claim in Greater Manchester? If not, what’s the next best in stock, and can we deliver within two days?”
The AI agent queries the relevant systems, returns the optimal answer, and actions the update directly. There is no UI, no form to complete, and zero training required to use five different software portals.
This isn’t five or ten years away. Nadella said recently that 2026 is the year AI stops being experimental – it’s now about real-world impact. Altman, on the BG2 podcast, talked about agents moving from multi-hour to multi-day tasks. If you’ve played with any serious voice-based agents recently, you’ll know it’s closer than most people think.
The Elephant in the Room: Trust and Guardrails
Of course, moving away from visual screens introduces a massive challenge: deterministic accuracy. If a technician simply talks to a system, how do we prevent misinterpretations, ensure strict regulatory compliance, and cleanly audit the results?
Shifting to a UI-less architecture doesn’t mean giving up operational control; it means changing how we enforce it. The core design challenge pivots from visual layout to systemic guardrails. We must build ironclad operating parameters ensuring that even when the visual interface disappears, the underlying logic remains perfectly transparent, auditable, and secure.
From Platform Thinking to Interaction Thinking
This completely changes how we think about building enterprise software. The future isn’t a suite of applications hidden behind a corporate portal; it is an intelligent layer of agentic capability that rides on top of core systems and responds to natural language.
For us at Bamboo Connect, that means:
Replacing portions of our manual claim flow with agentic voice execution.
Allowing clients to ask complex, real-time reporting questions without a dashboard.
Automating data collection and decision-making within our SLA operations.
Embedding trained, contextual agents directly into our call center stack.
Instead of designing for the UI, we need to start designing for understanding. We need to focus on how agents parse human intent, how they access background systems safely, how they know when to escalate to a human peer, and how they justify their actions.
Voice and autonomous agents won’t kill UX; they will simply give it a brand-new job. Designers will spend less time arguing over button placement and pixel alignment, and far more time figuring out how an agent should behave when it encounters uncertainty, or how it should clearly justify an operational decision to a regulator.
The End of UX as We Know It
We are transitioning from platform thinking to interaction thinking. This doesn’t mean the graphical user interface is entirely dead—there will always be a place for a well-designed chart or a visual data summary. But the era of “screen-first” software design is coming to a close.
The new frontier belongs to conversational execution for operational tasks, where design shifts away from layouts and moves toward protocols, prompts, and policy enforcement. A “good design decision” will become something as unglamorous—yet vital—as mandating that when a claim looks anomalous, the agent must always explain its reasoning before escalating to a manager.
As someone who has always loved great product design, I won’t miss the endless debates over pixel-perfect button placement. I’d rather describe the outcome and let an intelligent agent figure out how to get there. In the age of agentic AI, the best interface might not be visible at all – it just works.
About the author: Joy has spent more than two decades leading transformation across financial services, technology and operations, holding senior C-level roles and previously serving as a partner at Deloitte. His career has focused on scaling businesses, simplifying complexity and using technology to improve how organisations operate and grow.
Sources & InspirationRuccia, S.L. (2024). The Rise of Human-Grade Voice. https://shawnruccia.substack.com/p/human-grade-voice
Douglas, S. (2025). Why Claude Code is Utterly Wild.https://www.youtube.com/watch?v=mJnMcCNbEOs
Huang, J. (2025). The New Programming Language is Human.
Nielsen, J. (2024). AI: The First New UI Paradigm in 60 Years.
Nielsen, J. (2024). Superintelligence Will Kill UX.
Whittemore, N. (2025). The AI Daily Brief: Will Claude Code Kill the UI? https://podcastaddict.com/episode/214521142
Nadella, S. (2026). The Next Chapter in AI: Real-World Systems. https://cloudwars.com/cloud-wars-minute/satya-nadella-outlines-the-next-chapter-in-ai-real-world-systems/
Altman, S. & Nadella, S. (2025). BG2 Podcast: OpenAI & Microsoft. https://founderboat.com/interviews/2025-11-01-openai-sam-satya-microsoft/
Shah, D. (2025). Agents Are the New Apps. Agent.ai launch.https://www.agent.ai
Gartner. (2025). 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026. https://www.gartner.com/en/newsroom/press-releases/2025-08-26-gartner-predicts-40-percent-of-enterprise-apps-will-feature-task-specific-ai-agents-by-2026-up-from-less-than-5-percent-in-2025
Anthropic. (2025). Claude Code. https://www.anthropic.com/claude-code