Intention, Interfaces, and Agents: How I Think AI Is Influencing User Behavior

I'm interested in how AI changes the way people use software, alongside what the models can do. Discussions of large language models (LLMs), assistants often called copilots, task-performing agents, and Artificial General Intelligence (AGI), or broadly capable AI, tend to emphasize capability and its effects on work.

Those questions are useful, but the interface also shapes who can use a capability and how much work it takes to get a result.

Back when I was working my way through my undergrad degree in Professional & Technical Communications, I learned about media theorist Marshall McLuhan who said: “The medium is the message.”

I understood McLuhan's point as an invitation to examine how the structure of a medium shapes behavior, rather than focus only on its content.

That perspective makes the change in interaction interesting to me: people can increasingly describe what they want rather than operate every step of a digital system.

---

Technology Shifts When the Interface Shifts

Earlier changes offer ways to examine that relationship between a capability and how people encounter it.

The printing press changed both the production of books and the movement of knowledge. Information that had depended heavily on institutions, experts, and spoken communication could be distributed more broadly through print.

Television organized distribution differently, with networks such as NBC and CBS bringing centrally selected programming to millions of households. Its structure supported shared audiences and narratives.

The internet created a more distributed route for publishing and discovery.

Companies like Google and Facebook turned media into a networked system. Information became searchable, linkable, and personalized. Instead of a handful of broadcasters deciding what people saw, algorithms and social networks shaped content discovery.

Those examples show why examining the interface is useful alongside the technology: each changes how information moves and who can participate.

---

Software Has Always Been an Interface Problem

Early computers required humans to adapt to machines. Interaction happened through command lines and declarative instructions and users had to learn the system's language.

Personal computers changed the interaction model. Graphical interfaces let people navigate systems visually rather than programmatically. By using icons, windows, and menus, companies made software accessible to non-technical users. Apple and Microsoft built entire ecosystems around this, turning computing from a specialist skill into a mainstream one.

Mobile's introduction kept pushing the interface. With the iPhone in 2007, interaction became touch-first and always in reach. Software moved from desktops into pockets, and computing became embedded in everyday life. Software could run in the background whether people were consciously thinking about it or not.

Cloud software added another layer. Software as a service (SaaS), delivered online, connected work across teams, companies, and devices.

Throughout those changes, much of software use still involved navigating menus, clicking buttons, completing forms, and triggering processes. The user had to know enough about the system to operate the steps.

---

Copilots Are the First Sign of Something Different

Copilots offer another interaction pattern, where describing the task can become part of doing it.

GitHub Copilot, Microsoft Copilot, and ChatGPT let users describe an intended result and receive help producing it, rather than specify every operation through the conventional interface.

Much of that use still happens within existing workflows, helping with text, code, summaries, and questions without removing the underlying software. The interaction sits between direct operation and broader delegation.

---

Agentic Systems Change the Relationship

An agentic model extends that relationship into delegated work. A user describes an outcome and gives the system authority to perform appropriate steps, rather than only ask for information or content.

A conventional analysis workflow might involve:

1. Open the system
2. Navigate to the right page
3. Export data
4. Clean it
5. Run analysis
6. Create a report

An agentic interface aims to compress that chain into a single request: "Analyze last quarter's churn and show the biggest drivers."

Here, churn means customers leaving or stopping their use of a service. The user describes the question to answer rather than operating each step of the workflow.

---

Early use doesn't establish mainstream behavior

The changes are visible in engineering, product building, and AI research, where people use assistance for coding, debugging, research, and design. I've seen and heard of developers treating the tools as collaborators in that process.

Those early users may be comfortable with models, prompts, workflows, and coordinated agents in ways other users aren't. Someone running a business may have little time or reason to learn the mechanics; their purpose is completing the task.

If AI becomes more deeply embedded in systems, some mechanics may become less visible. Whether that helps depends on how well the interaction supports the user's work, not just whether it feels new.

---

When the Medium Disappears

It's possible that AI follows technologies that became ordinary parts of the operating environment rather than remain a distinct category users think about continually.

Electricity became infrastructure. The internet became a background layer across communication, commerce, and media. Smartphones became extensions of everyday life.

AI may similarly become part of operating systems, productivity tools, search, customer interfaces, creative work, and commerce.

People might then describe the task rather than think of themselves as "using AI." That remains a possible direction, not an assumption about every user or system.

---

What This Means for Small Businesses

An interface can expand access to an existing capability by reducing what a person must learn to use it.

Printing changed the distribution of knowledge, the internet made publishing and reaching audiences more accessible, and mobile devices made digital commerce available in more situations.

AI may reduce some of the effort of using software in a comparable way.

For small businesses, analytics, marketing automation, customer relationship management (CRM), and operations platforms already offer capabilities that take time to configure and learn. A smaller organization may not have a dedicated person to manage each tool.

An interface that understands a request could reduce some of that navigation and configuration work. The person could describe what they need to learn:

"Show me which customers stopped ordering in the last 90 days." "Draft a follow-up email for recent buyers." "Create a report showing which products drive repeat purchases."

That could let someone running a 10-person operation ask questions of the data without first learning an analyst's software workflow. It doesn't remove the need to understand whether the information and answer support a sound decision, but it may make the capability easier to reach.

---

Designing for a Different Relationship

Designing that interaction means deciding how authority and information move between the user and system.

Traditional design helps people navigate and complete workflows. If software also interprets intent and performs actions, the design needs to establish which parts of the process a person directs and which can be delegated.

Useful questions include which outcomes people can describe clearly, which tasks the system can perform appropriately, and when it should act or ask. Those decisions depend on context and consequences as well as the quality of the interface.

Product teams can use that understanding to choose where assistance reduces effort and where the user's judgment needs to remain explicit.

I think the shift toward collaborating with software is worth examining separately from increases in model capability. It changes the relationship between people, tools, and the work, which can affect how the wider system operates.