Technology Design in the Age of AI
I’m passionate about design. I see beauty and genius in discovering novel structures and patterns that are functional and elegant in their simplicity. So I’m surprised more people are not talking about AI disrupting over 75 years of computer-related design. To me, it seems like an amazing inflection point.
Current Design Context
The term “computer” was originally assigned to human beings – in particular, individuals that compute things for others. Acknowledging the routine work associated with computing answers, Charles Babbage developed the concept of a mechanical programmable computer (some people may recall that GameStop’s original name was Babbage’s). Electromechanical computers followed in the 1930s-1940s, with Colossus and ENIAC becoming the first electronic digital programmable computers. Through the rise of transistors, integrated circuits, and today’s most powerful processors and architectures, one design principle has held constant: humans would be the main users of this technology.
But AI is challenging long-standing hardware and software architectural and design concepts. I’ve compiled a list of the biggest design-related shifts I see emerging from the evolving capabilities in AI.
Computers as First-Class Users
To many people, the most fundamental and obvious shift broadens the intended developer and user community from humans to include machines. Though information systems have participated in workflows for decades, AI-empowered systems feel like a different design metaphor.
AI is a primary developer and user. Software will no longer primarily be built just for people to click and type. Instead, AI programs themselves are becoming priority users of digital systems, fundamentally changing how everything is designed from the ground up. Systems will become optimized for autonomous AI interactions, and new software applications will be built by software.
AI becomes inseparable from software engineering. The process of building software is being transformed by AI. AI tools will assist developers in generating code, creating comprehensive test suites, discovering obscure edge cases, and predicting potential build failures. The role of human software developers will evolve from writing every line of code to becoming guides and supervisors for AI tools, focusing on designing the overall system, defining complex problems, and overseeing AI-generated solutions. This of course changes the design of integrated development environments (IDEs).
Autonomous AI agents perform routine work. Future software will be powered by intelligent programs, or AI agents, that can observe, reason, and act independently. These agents can plan their own steps and execute complex tasks without needing constant human oversight. Designs will need to reflect this new operating autonomy. And AI will move into IT operations, leading to self-healing systems. AIOps platforms will use machine learning to monitor software performance in real-time, predict potential issues, guard against cybersecurity threats, initiate maintenance, and even automatically resolve problems without human intervention.
Data gets designed and aggregated for machines: The way information is stored and organized is changing. Data architectures will be built considering the need for AI to process and interpret data efficiently, rather than just for humans to read or analyze. Vector databases and other emerging models allow AI to find things based on their meaning and similarity, not just exact keywords.
Software Value Flows from Data
For most of the digital computing age, software value derived from the code – the algorithms and associated functionality they deliver. But as code becomes an automated commodity, AI considerably alters that value equation:
Data is a differentiator. General-purpose AI models are rapidly becoming a commodity. But specialized models – trained and tuned to specific tasks using highly curated data – are not a commodity. And that means organizations that invest in these data assets – and the software solutions that generate that data – will enjoy more upside.
Data timeliness gains prominence. AI models will increasingly rely on diverse methods of perceiving the world around them, with data flowing continuously and instantly throughout software systems. Robotics and sensors will broaden the focus from camera and microphone inputs to include multispectral vision, spatial audio, haptics, balance, position, velocity, olfaction, bioelectric, chemical, energy, gravimetric, telemetry, and even internal state.
Software Becomes More Human & Dynamic
There was a time when technologists needed a lot of user manuals. Those days are coming to an end as software archetypes reflect a bias towards human interactions.
Conversational interactions drive information delivery. Instead of navigating menus or learning complex commands, users will increasingly talk or type naturally to software. Interacting with systems becomes less about functional mastery (i.e., knowing the software) and more about subject mastery (i.e., understanding what needs to be done).
Speech becomes a preferred I/O model. Because application functionality will be delivered through conversational exchanges, it will seem more natural and expedient to simply talk to some AI models and agents. This voice mode option means that AI interactions will occur in times and places where typing and screens would have limited convenient use of earlier AI systems.
Intent-driven software will displace many linear applications. The software of the future won't just follow step-by-step instructions. It will be designed to understand high-level goals and intentions, then figure out the best way to achieve them. This means users will tell the system what needs to be done and why, but not how to do it.
Application functionality will be built on-the-fly. Historically, application functionality is designed, built, and tested as part of a product R&D cycle. Though that will continue, functionality will also be dynamically coded on demand by AI models. The models will discern the user’s request, build a tiny program to deliver it, execute the program, and return the results.
Contextualized interfaces replace one-size-fits-all user experiences.Software user interfaces have gradually been getting smarter for decades, shifting from a static set of menus to context-aware displays and functionality. AI will push that trend even further with interfaces that are generated dynamically according to individual needs, preferences, and the activity currently underway.
Integration As A Primary Design Criterion
Most organizations will see AI as a means to automate workflow across systems. As such, the ability to connect differing systems and data together moves from a “feature” to a critical design element that drives value delivery from AI models.
Integrations are built for and with AI. The connections between different software systems (e.g., APIs and other data interchange functionality, protocols, and standards) will be designed specifically for intelligent AI agents. These mechanisms will allow AI programs to communicate more effectively, maintain context across interactions, and use tools more flexibly. To ensure different AI models and systems can work together seamlessly, new industry standards and protocols will emerge that define how AI agents communicate capabilities, exchange context, and integrate across various platforms.
Retrieval-oriented AI designs will be everywhere. Since AI models need access to enterprise data repositories for processing (e.g., RAG), solution architectures that explicitly incorporate retrieval capabilities will be a common practice. Enterprise systems, AI agents, MCP servers, and workflows will all incorporate this design requirement.
Modular AI microservices deliver extensible functionality. Instead of one large program, software will be broken down into many small, independent AI capabilities that operate as separate services. This microservices approach allows each AI function to be developed, scaled, and updated independently, making systems more flexible and robust.
Computing Paradigms Expand
From its origins as room-sized machines through the handheld computing revolution and wearables, information technologies have become more pervasive and accessible. AI will further amplify this perfusion of technology.
Hardware options will further diversify. Because many AI capabilities will be cloud-resident, a broader array of hardware form factors will be able to use them. Typical constraints – screen and keyboard access, for example – become less important than connectivity to networks and sensors that can feed data to models and rapidly interpret responses.
AI operating systems will emerge. Historically, operating systems have existed to enable user access and control of hardware. With the shift in focus from human to machine users, operating systems will evolve from basic hardware control systems and user-facing features to agent and task control systems designed to support autonomous applications.
Many AI deployments will be services, not servers. Many AI tasks, especially those that run occasionally or have unpredictable demand, will use serverless computing. This means the cloud provider automatically manages the underlying infrastructure, and users pay for compute time. This approach changes the economics of many corporate IT strategies.
AI deployments will also be device-resident. As AI methods evolve and mature, many AI models will be small, capable of being deployed within individual devices. Localized rapid processing, especially in fields such as computer vision and language processing, help ensure low latency responses and resilience to network disruptions.
Event-driven designs drive scalability. Software components will communicate by reacting to "events"—significant changes in the system's state. This event-driven architecture enables AI agents to respond instantly to business events, triggering autonomous actions and making the entire system more responsive and scalable.
Safety as a Design Priority
Other sectors – automotive, children’s toys, and more – have long embraced safety as a critical element of product design. Beyond industry-specific use cases, historical software designs have not needed to consistently embrace a safety mandate, but it will be non-negotiable for many AI technologies, whether stipulated in regulations or not.
Models will differentiate on trust. To ensure high quality and drive adoption, some AI systems will actively aim for explainability (understanding why AI made a decision), fairness (avoiding bias), rigor (ensuring all steps were taken), and reliability. These models will be favored in some mission-critical settings where process adherence ensures quality and safety.
New AI security models exist alongside human-oriented security practices.Security systems will recognize AI agents as distinct, non-human entities. This advancement requires specialized ways to authenticate and manage AI access, as traditional human-centric security methods like multi-factor authentication don't apply as easily to machines.
Secure-by-design will increase for AI solutions. Security will shift to address new, unique risks posed by AI, such as "prompt injection" where attackers trick AI models, or vulnerabilities in AI-generated code. Future software will have built-in defenses specifically designed to protect against these AI-specific threats.
Designs Must Account for Faster Innovation
Today, there doesn’t appear to be a speed limit on AI advancements. Given the pace, designs will need to be both innovative and adaptable – a hard design goal to meet.
Generative AI will further diversify beyond existing LLMs. Though the current generation of large language models are driving many digital transformation programs, the universe of AI models and methods – small language models (SLMs), state-space models (SSMs), mixture-of-experts (MoE) architectures, diffusion models, hybrids and others – will increasingly demonstrate a more diverse, extensible, and fit-to-task array of AI options.
AI walled gardens compromise design goals. As AI manufacturers move to accelerate adoption with their own features, cross-vendor compatibility suffers. The IBM PC offered a pathway to enabling multiple hardware manufacturers to participate in the desktop computer revolution. For comparable interoperability to occur with AI, vendors will need to move beyond proprietary SDKs, methods, and function calls.
Quantum computing will open a parallel branch of AI. Though quantum computing resources will not be as accessible as conventional computing architectures, its massive scale in capacity, complexity, and performance will usher in an alternate path to developing and exploiting some AI capabilities.
Planning Ahead
Can AI generate the power, passion, and creativity inherent in great human designs? Certainly in areas where designs emerge from computational tasks (e.g., derivative designs, modelling, simulation, optimization), it seems likely. We often appreciate design as an art; the Porsche 911, the Lamy 2000 fountain pen, and the Eames Lounge Chair persist because they beautifully blend function and form. Similarly, many current IT architecture concepts – service orientation, statelessness, separation of concerns– will undoubtedly propagate into the future as well because they work well.
But design is also alive. Electric vehicles have prompted automotive companies to re-assess design elements like drive trains, battery placement, controls, and trunks. AI is having a similarly disruptive impact on the way we think about our information technology. Though legacy designs are not disappearing tomorrow, organizations should be developing strategies (business, IT, and AI) today that establish the roadmap for embracing these ongoing shifts. Many of these design shifts are already here, and others are arriving faster than many anticipate.
