About this video

• Video Title: Agents Will Kill Your UI by 2026--Unless You Build This Instead
• Channel: AI News & Strategy Daily | Nate B Jones
• Speakers: Nate B Jones
• Duration: 26:13

Overview

This video discusses the fundamental shift occurring in software development and user interfaces due to the rise of generative AI and agentic software. The speaker argues that traditional, hand-crafted UIs are becoming obsolete ("throwaway pixels") and that software will increasingly be generated on demand based on user intent and context. The video explores the implications of this change for B2B SaaS, talent allocation, and the future of software design.

Key takeaways

• The Decline of Coherent Interfaces: For decades, user interfaces (UIs) were expensive to design and build, leading to durable, shared interfaces. This era is ending as generative AI makes UIs cheap and contextual.
• Three Overlapping Shifts: The transition is driven by generative UIs that can create full screens from prompts, the explosion of ephemeral and contextual UI concepts, and the rise of agentic software that drives other software.
• "Disposable Pixels" vs. "Substrate": Software is decoupling into a stable "substrate" (data models, workflows, logic) and "disposable pixels" (UIs generated on demand). The core value will shift to the substrate and agentic capabilities.
• Impact on B2B SaaS: Companies that rely solely on their UI for differentiation are at risk. Success will depend on being a "substrate as a service" with clean schemas, agent addressability, and composability.
• Talent Reorientation: Designers will become "language designers and safety engineers for human attention," PMs will focus on supporting intents and safe state changes, and engineers will build stable interfaces for agents.
• Enduring Coherence: While disposable pixels will dominate, coherent interfaces will still exist for specific use cases like regulated environments, complex cognitive tasks (e.g., trading), and deep team collaboration where stable landmarks are crucial.
• The Future is a Spectrum: Software will likely exist on a spectrum, with highly standardized, coherent shells for certain functions and ephemeral, disposable pixels for exploratory and personalized tasks.

The three overlapping shifts driving the change in software interfaces are:

1. Generative User Interfaces: These are models capable of producing entire screens from text or contextual prompts. Examples include UISard, Vzero, and Galileo, which can generate multi-screen mock-ups from prompts. This leads to the concept of "cheap disposable UI."
2. Exploding Ephemeral and Generative UI Concepts: There's a growing focus on hyper-contextual applications and panels that can be created and destroyed, maintaining application state. This signifies a shift towards a new design language for these dynamic interfaces.
3. Agentic Software: This refers to software that drives other software. Tools like Nano Banana Pro, when exposed via APIs, can be called by agents to perform tasks such as generating charts and graphs based on structured prompts, enabling generative interface design driven by agentic software.

The speaker differentiates between "disposable pixels" and the "substrate" in software as follows:

• Substrate: This layer is described as the durable foundation of software. It includes components like data models, workflows, permissions, audits, compliance rules, domain logic, forecasting engines, pricing engines, APIs, and webhooks. The speaker emphasizes that this layer is highly valuable, dense, and enduring; it's not going anywhere and is where moats are built. For B2B SaaS, this represents the core system of record or decisioning.
• Disposable Pixels: These are the user interface elements that are generated on demand and are contextual to the user's immediate needs. Unlike the traditional, expensive-to-build, and durable interfaces, these pixels are considered "cheap" and ephemeral. They are compiled artifacts of intent, appearing only when needed for a specific decision or task. This layer is transient, specific to a moment or query, and can be discarded once its purpose is served. This contrasts with the traditional "learn this app" mental model, moving towards a "state your intent, do the prompt, and UI appears" paradigm.

The shift towards disposable pixels and agentic software has significant implications for B2B SaaS companies. The speaker highlights that companies whose primary value proposition is their user interface are at risk.

Here are the key implications and characteristics for future success:

Implications:

• UI is No Longer Fully Controlled: When customers use generative AI tools on top of a SaaS company's APIs, they can create their own views of the data using their own design systems. The SaaS provider's native UI becomes just one of many potential frontends.
• Risk of Disintermediation: SaaS products can become aggregated behind a single agentic interface, potentially weakening the direct relationship between the vendor and the customer.
• Shift in Competitive Landscape: The focus shifts from having the "best dashboard" to being the "easiest system for agents to choreograph."
• Value Moves Up the Stack: Enterprise value is currently framed around owning the primary surface where work happens. If this interaction moves to an agent or co-pilot, the SaaS UI becomes a reference implementation rather than the default touchpoint. API behavior and data semantics become more critical than the navigation bar.

Characteristics for Future Success:

• Be a Substrate as a Service: The core value will lie in being a stable, reliable foundation. This means owning the canonical state for critical business functions like contracts, ledgers, records, and risk models.
• Agent Addressability: Products must be easily accessible and callable by AI agents.
• Schema Cleanliness: Having well-defined and clean data schemas is crucial for agents to reliably interact with the software.
• Composability: Software needs to be built in a way that allows different components and agents to interact and build upon each other.
• High Integrity Service: Instead of just being a "thing with screens," companies need to be a reliable service that agents and generators can depend on. This includes strong schemas, good safeguards, and item potency.
• Embedded in Domain Flows: Products that are integral to core business processes and track real value (like SLAs, compliance, reference data) will remain essential.
• Treat UI as Language and Runtime: Teams should view UI not as a set of static screens but as a dynamic language and runtime environment that can adapt.

In essence, B2B SaaS companies need to focus on the robustness, data integrity, and API-driven capabilities of their underlying platform rather than solely on the aesthetics or functionality of their front-end interface.

The speaker outlines significant shifts for designers, product managers, and engineers in the era of AI-driven interfaces:

For Designers:

• From Owning Flows to Defining Grammar: Designers will move away from owning specific screens and flows. Instead, they will focus on defining the "interface grammar" – the constraints, safe interaction points, and rules for generative UI.
• Becoming Language Designers and Safety Engineers: Their role will evolve into designing the language that AI agents use to create interfaces and ensuring the safety of human attention within these dynamic systems.

For Product Managers (PMs):

• From "What Feature Next?" to "What Intents to Support?": The core question will shift from deciding which feature or page to build next to identifying what user intents the system should support.
• Focus on State and Automation: PMs will concentrate on defining what state changes need to be safe and which decisions require human judgment versus full automation.
• Specifying Intent, State, and Outcome Loops: Instead of creating static wireframes, PMs will be responsible for specifying complex intent, state management, and outcome loops for AI agents.

For Engineers (especially Front-End):

• From Pixel Pushing to Agent/Generator Interfaces: Front-end engineers will transition from meticulously crafting pixels for direct human interaction to building stable interfaces that agents and generators can interact with.
• Building Stable Interfaces and Shells: They will need to develop a "thin canonical shell" that enables snap points for generative UI, validation logic, and a degree of composability within safe constraints.
• Updating the Backlog: The backlog will change from traditional tickets for new pages to new types of work, such as supporting new intents, defining system behaviors, establishing new constraints or invariants, and creating components or layouts that generators can utilize.