CIOs reimagine software’s future as AI agents advance

Anthropic’s inconspicuous release of software plugins in January triggered a furious sell-off of software-as-a-service (SaaS) stocks. Over the next two weeks, the financial markets watched $300 billion in software valuations vanish.

Anthropic’s Claude Cowork brings a more user–friendly version of the company’s developer-focused Claude Code to business users. The software enables knowledge workers to prompt an AI agent to organize files, create documents, and more. Plugins for Cowork bundle skills, connectors, and subagents aimed at specific business roles, seemingly to automate the knowledge work that fuels billions of dollars in recurring SaaS revenue.

Those who tested them quickly concluded that SaaS is in trouble and the SaaSpocalypse was born. But the frenzy subsided, the market calmed, and sanity was restored. SaaS is not dead or dying.

Even so, the events raised the specter of what the future of software will look like as organizations more heavily embrace AI agents and generative AI tools to automate business processes.

AI fuels workflow changes

As they embrace AI more tightly, IT leaders are changing everything from workflows and user experiences to operating models.

Under Chief Digital and Technology Officer Greg Meyers, Bristol Myers Squibb used AI to overhaul a critical gross-to-net forecasting system, which translates medicine prices into revenue by accounting for global pricing, regulations, and international market rules.

Rather than optimize an existing solution that relied heavily on spreadsheets and enterprise systems, Meyers’ team seized on the speed and capabilities of AI to rebuild the software. The resulting solution cut forecasting errors by 50%.

The AI era presents organizations with an opportunity to approach a lot of your “fundamental business processes with a clean-sheet of paper,” Meyers tells CIO.com.

Meanwhile Johnson & Johnson is using AI to reimagine workflows — with a human always in the loop — which has been a “game-changer,” according to CIO Jim Swanson.

The company is using GitHub Copilot to assist in software development from user stories to testing and validation, which Swanson says is critical for a pharmaceutical company operating in a regulated industry.

IT is also using AI to automate functions such as resolving customer service issues, which will drastically reduce the number of screens human employees must access as they execute their tasks.

Software is in for a UX overhaul — but not overnight

Swanson’s point about reducing screens underscores where software is headed in the future.

Today human employees still go bleary-eyed clicking from one UI to another, and from one application to another. Eventually, how employees interact with AI-infused software systems will change, says Arnal Dayaratna, research vice president of software development at IDC.

As agents become more embedded in software and workflows, coworking alongside humans the bots will handle more of this work, including filing work orders, service tickets, and other executables. Agents will communicate via APIs, much the way disparate applications interoperate today.

Humans will oversee agent execution through command-and-control dashboards. For instance, these screens will show whether agents are performing as intended and flag when they are not.

When they fail to stay on task, human employees will correct them through a chat interface similar to the way they prompt generative AI applications powered by LLMs today, or even through voice interfaces as the world shifts toward more multimodal AI interactions, Dayaratna predicts.

Agentic AI hurdles remain huge

There are myriad reasons to pump the brakes on AI autonomy. Agents remain brittle, challenging accuracy and reliability. Tech leaders are well aware of this, dismissing the notion of full agentic autonomy in their organizations — at least, for the foreseeable future.

For instance, customers are leery of pushing a button triggering software to proactively remediate issues with PCs, notes Fasul Masud, president of HP’s worldwide digital and lifecycle services.

“Why do we think a CIO is going to enable multiple thousands of agents to make decisions and direct those agents through other agents without having governance on top?” Masud asks. “IT leaders won’t expose tier-1 systems to autonomy.”

Swanson confirms that position: “I’m not going to have 10,000 agents running in the environment that I don’t know what they’re doing,” Swanson says. “There’s no company that can accept that.”

Masud adds that while it’s increasingly unlikely humans will have “eyes on glass all day,” employees will still need to review reports to ensure quality control of deliverables. In short, humans’ work is shifting from tactical and technical to governance tasks.

Moreover, Dayaratna cautions that the industry must make significant progress in improving the discoverability, orchestration, and collaboration between agents.

“How does an agent know when another agent’s work is done?” Dayaratna asks. “How does the testing agent know that the code is ready to be tested?”

The answers to these critical questions will determine when agents are ready for enterprise prime time.

SaaS will be fine — while software complexity balloons

As to the future of SaaS, experts believe the huge category will continue to coexist — robustly, at that — with agents. In fact, SaaS vendors are already embedding specially crafted agents within their applications, proving that it’s not either SaaS or agents, but SaaS with agents.

“’Death of SaaS’ narratives are overstated,” notes Forrester analyst Kate Leggett. “The brain of the enterprise remains, the central nervous system is evolving, and the center of gravity is becoming more intelligent.”

Some believe SaaS may be poised for more growth with the rise of agents. As SaaS vendors pointed out, organizations will need to build more software for agentic guardrails.

The potential downside of SaaS-plus-agents-plus-genAI-plus-on-premises apps? Greater software complexity than exists in the enterprise world today, setting the stage for more technical debt than even the most cynical software engineer can imagine.

No one knows just how deep that rabbit hole goes.