It’s 8:45 on a Tuesday morning in 2029, and you’re running late for your first call. You haven’t logged into your customer relationship management system in months. You don’t need to. Your digital sales agent has already triaged the overnight leads, selected the ones that merit your attention, and scheduled two that look particularly promising.

By the time you finish your coffee, another agent has pulled the latest pricing data from finance, checked inventory against the supply chain system and drafted a personalized proposal for your noon meeting. Meanwhile, a compliance agent in the background has double-checked that all documents adhere to regulations. You didn’t ask for any of this; the agents simply know your goals and orchestrate the workflows accordingly.

Your workday is no longer about using applications. It’s about collaborating with a network of artificial intelligence colleagues who anticipate your intent, connect across silos and adapt in real time. AI agents can reason, make decisions, suggest improvements and take actions, making them far more powerful than question-answering large language models.

Intent and outcomes

“We are evolving into systems of intent and outcome where the back end will be able to derive the intent of the users and get the work done,” said Arin Bhowmick, chief design officer at SAP SE.

The shift from static software to agentic ecosystems isn’t science fiction. It’s the next logical step in the evolution of enterprise technology, and it’s happening faster than many people realize.

SAP’s Bhowmick: “We are evolving into systems of intent and outcome.” Photo: SAP

Enterprise software vendors are scrambling to embed agents into existing applications. Oracle Corp. claims to have more than 600 embedded AI agents in its Fusion Cloud and Industry Applications. SAP says it has more than 40. Salesforce Inc. is rebranding its entire product line around agents with the recent launch of Agentforce 360, a platform that infuses agents into nearly every application the company delivers.

“In the agentic world, your capabilities are essentially anything that has any kind of endpoint on the internet,” said Nancy Xu, vice president of AI and Agentforce.

The building blocks are already in place. Enterprise applications are evolving from siloed tools into interconnected, agent-driven ecosystems that collaborate with humans and each other to get work done.

“You can think of agents as the new middleware,” said Federico Toretti, senior director of AI and machine learning at Oracle. “They’ll do the work of connecting, contextualizing and coordinating across applications.”

Rethinking software

This shift is not simply about embedding AI into existing products, as generative AI is supplanting conventional menus and dashboards. It’s a rethinking of software’s core functions. Many experts working on the agentic future say the way software is built, packaged and used is about to change profoundly. Instead of being a set of buttons and screens, software will become a collaborator that interprets goals, orchestrates processes, adapts in real time and anticipates what users need based on their behavior and implied preferences.

Salesforce’s Xu: “Depending on how you interact, the UI evolves with you.” Photo: SiliconANGLE

Generative AI interfaces are already making software more personal and adaptive. For decades, enterprise applications have forced users to adapt to working their way through dropdown menus, fixed fields and minimally configurable dashboards. Generative AI is now replacing many of those awkward constructs with natural language commands.

AI agents flip the model even further by learning from individual behaviors, preferences and context to shape experiences in real time. Instead of issuing commands, users respond to prompts and suggestions from the machine.

SAP’s Bhowmick calls this shift hyper-personalization. “Traditional interfaces are created with a specific idea of a persona; hyper-personalization is the other way around,” he said. “The system adapts to the human.”

That means that a manager who prefers conversational interfaces might get options presented in a natural-language dialog, while power users see code. Or agentic systems can surface just-in-time insights without requiring a query or prompt simply because something has changed that might be meaningful.

“We’re moving toward anticipatory and adaptive experiences,” said Jeff Gelfuso, chief product and experience officer at Qualtrics Inc. “Agents will anticipate that you and I would have a conversation on these types of topics and will bring forth things that prepare us for a meeting. It might also give us answers after the meeting, and opportunities to follow up.”

Salesforce is adapting the user experience across its portfolio to emphasize this personalization, Xu said. “Every customer is getting a completely different UI experience,” she said. “Depending on how you interact, the UI evolves with you.”

UI for you

Oracle’s Toretti foresees an end to static user interfaces. “In the future, no two employees will experience the same software in the same way,” he said. ”The interface will reshape how individuals work.”

Adaptive UIs reduce training time, lower frustration and enable more inclusive access. They also turn user experience into a competitive differentiator. Software firms will increasingly compete to offer experiences that integrate seamlessly into each user’s workflow and adapt to a company’s brand and governance settings.

“The future is software that learns your patterns but respects your autonomy, such as showing procurement officers supplier risk scores during negotiations while presenting warehouse managers with capacity optimization suggestions,” said Deepak Singh, chief innovation officer at Adeptia Inc., a maker of an AI-powered platform that connects, integrates and exchanges data between different systems without coding.

Shift to modularity

Oracle’s Toretti: “In the future, no two employees will experience the same software in the same way.” Photo: LinkedIn

The coming changes to enterprise software will go beyond the interface. AI will force monolithic software stacks to give way to modular, composable systems stitched together by agents using standards such as the Model Control Protocol, the Agent2Agent Protocol and the Agent Communication Protocol that IBM Corp. recently donated to the Linux Foundation.

“By 2028, AI agent ecosystems will enable networks of specialized agents to dynamically collaborate across multiple applications, allowing users to achieve goals without interacting with each application individually,” Gartner recently predicted.

“Imagine a supervisor agent that talks to one of Starburst’s built-in agents, but also can talk to Tableau or Power BI,” said Matt Fuller, co-founder of Starburst Data Inc., maker of a data lakehouse platform. “The agent helps orchestrate the end-to-end experience from accessing the data and preparing it to create dashboards downstream.”

That has profound implications for how software is constructed and packaged. Software has traditionally been self-contained, with all functionality packed into a single codebase under the developer’s control.

That model began to break apart with the shift toward cloud computing and software-as-a-service, which exposes services via application programming interfaces that can be called as needed. Extensibility is one of the chief attractions of cloud-native computing.

Protocols such as MCP and A2A extend this metaphor to intelligent services that can swap tasks. Agents’ ability to reason, perform tasks and invent workflows will make the resulting applications far more sophisticated than what can be achieved with APIs alone.

“Every potential MCP server in the world becomes part of the functionality of your agent,” said Salesforce’s Xu.

The new marketplace

Qualtrics’ Gelfuso drew an analogy to the marketplaces that sprung up around smartphones and SaaS. “Monolithic software has been broken down into much smaller purpose-built applications in those marketplaces,” he said. “I think you’ll see a very similar type of trend here.”

Qualtrics’ Gelfuso sees monolithic software giving way to specialized components. Photo: LinkedIn

In that scenario, the role of software companies is less about addressing soup-to-nuts needs than building core functionality that is easily and reliably extended. Ecosystems and low-code/no-code toolsets will be critical competitive factors. The software giants of the future could be those that are most effective at orchestrating armies of agents.

“We’re moving past the notion of linear workflows into something much more about the outcome,” Gelfuso said. “How do I design an experience that will evolve with the model or the technology and still deliver that outcome?”

Some believe future software innovation will come less from commercial vendors than from their customers. The ability of agents to act autonomously gives users powerful control to define their own functions and workflows. Though “vibe coding” has failed to live up to its hype as a replacement for traditional software development, it has created an important new metaphor for prototyping and set the stage for nontechnical users to dictate how software will perform.

“The subject-matter expertise required to build agents oftentimes lives in your business line managers or nontechnical experts,” said Salesforce’s Xu. “We need to help them participate in the agent-building process.”

That potentially changes the role of software platforms from applications to frameworks that coordinate and govern agentic behavior. “The roles of software companies are going to be to enable a control plane where humans can orchestrate agents,” said Oracle’s Toretti. Software “evolves into a citizen that allows individuals to provide oversight that validates results or identifies errors and enforces guardrails.”

Fast and fluid

An agentic future holds out the promise that software will be built more rapidly and behave more fluidly. “The components are going to be created on the fly based on the type of question you are asking,” said Francois Lopitaux, senior vice president of product management at ThoughtSpot Inc. “All the cumbersome work around integration will be simplified. They are going to be able to exchange information and run workflows across applications.”

Starburst’s Fuller: Agents will “orchestrate the end-to-end experience from accessing the data and preparing it to create dashboards.” Photo: LinkedIn

Agents could also shake up the way software is packaged and paid for. Special-purpose agents that are called upon only occasionally to perform a single task are unlikely to command subscription fees. New pricing constructs will likely emerge based on value.

In May, Salesforce introduced Flex Credits, a consumption-based pricing model for its Agentforce services that has customers paying for specific actions performed by AI agents, such as updating records or automating workflows.

“Enterprise software will be sold by value, and what process problems can be solved… over simply buying an acronym,” wrote Larry Dignan, editor in chief of Constellation Research Inc.’s Constellation Insights.

But getting there won’t be fast, simple — or even possible. Enterprises rely on software to be deterministic, meaning that the same set of inputs will yield the same outputs every time. AI models are, by their nature, probabilistic. They estimate the probability of different possible outputs and choose those with the highest likelihood of success.

Probabilistic reasoning yields creative ideas, but that isn’t what a chief financial officer is looking for when closing the quarterly books. As agents proliferate, orchestration, guardrails and trust frameworks will be critical to ensure security, reliability and compliance.

“We will have the concept of governance agents… to make sure things follow protocols, are compliant, and apply ethics to what agents do,” said SAP’s Bhowmick.

Governance mandate

Plenty of work is going on to make agentic interactions more predictable. Salesforce’s Trust Layer covers tasks such as preventing sensitive data from leaking to AI models, governing which models can be used, checking output, and creating audit trails. “It enables customers to build agents in a more trusted and contained environment,” Xu said.

ThoughtSpot’s Lopitaux: “All the cumbersome work around integration will be simplified.” Photo: ThoughtSpot

IBM’s In-Context Explainability 360 Toolkit helps practitioners understand why an AI model made a specific decision within the context of the application where it’s being used.  “Some people claim that you can just ask the model about its reasoning, but there can be hallucinations in those explanations, too,” said Kush Varshney, an IBM Fellow based at the Thomas J. Watson Research Center. “We take a very mathematical approach.”

Numerous other initiatives are underway at the national, industry and university level. Adeptia’s Singh believes most of the governance issues will be worked out over time in the same way that APIs has become more robust and secure as they went mainstreat.

“APIs were initially not so secure,” he said. “They didn’t have quality-of-service features, rate limiting and security credentials. All of those capabilities will come to agents as well.”

Lack of predictability may ultimately be the biggest factor that holds agents back from remaking conventional software. Though some people have speculated that agents could make enterprise applications as we know them obsolete, experts who were interviewed said that is unlikely.

“Systems of record will have to be there because they are necessary in a lot of regulated industries,” said IBM’s Varshney.

Citing the “meaningful complexity” of enterprise workflows, Jefferies LLC Software Analyst Brent Thill recently wrote that “the intricacies of enterprise architecture make full AI disintermediation of software unlikely.”

Though systems of record may not go away, the ways we interact with them are likely to change. “Traditional transactional interfaces with forms and tables and dashboards will coexist, but the composition of the interaction layer will change a lot,” said SAP’s Bhowmick. “You will ask and interact with AI at any time, regardless of which app or app boundary you are using.”

That has the potential to effect the most momentous changes in the way organizations use software since the dawn of the commercial software industry nearly 60 years ago.

“You’re moving past the notion of linear workflows into something that’s much more about the outcomes you’re trying to achieve,” said Qualtrics’ Gelfuso. “We can all make predictions, but it’s changing so fast that I’m just thrilled to be a part of it.”

Image: SiliconANGLE/Google Gemini