From chatbots to colleagues: How agentic AI is redefining enterprise automation

Chatbots once symbolized digital transformation — those polite text boxes on corporate websites and service portals promised to make support smarter and cheaper. The addition of generative AI (genAI) to the tools in recent years has made them seem more natural in conversations, but they’re still just automated answer engines. Now as 2025 comes to an end, traditional chatbots are beginning to look like relics of an earlier era.

A new wave of agentic AI is taking shape: systems that not only converse but also reason, plan, and act within enterprise workflows. These agents are not assistants that talk; they are digital colleagues that think.

Across industries, companies are reengineering their operations to harness this new capability. They’re discovering that agentic AI isn’t simply an upgrade to chatbots — it’s a redefinition of how digital work gets done.

From reactive bots to proactive partners

Jesse Flores, founder and CEO of web development firm SuperWebPros, has watched this transition unfold firsthand. “Traditional chatbots,” he said, “were basically decision trees — if keyword X, then response Y.” They worked well for FAQs and appointment scheduling, but their world was bounded by the script.

Even when connected to large language models such as GPT-5, most chatbots still lack deep knowledge of a company’s data or business context. “They’re language-driven responders,” Flores explained. “They talk, but they don’t think or act.”

According to Flores, agentic AI changes that equation. Each agent has a name, a mission defined by its system prompt, and a connection to company data through retrieval-augmented generation. Many of them also wield tools such as CRMs, databases, or workflow platforms. “An agent is like hiring a new employee who already knows your systems on day one,” Flores said. “It doesn’t just respond — it executes.”

This new mode of collaboration also changes how employees interact with technology. Flores noted that his clients often name their agents, treating them as teammates rather than tools. “When marketing needs to check something, they’ll say, ‘Let’s ask Marco,’” he added. “That naming makes adoption easier — it feels human.”

Other insights can be gleaned from Moody’s, an organization well known for its information services. According to Cristina Pieretti, head of digital content and innovation at Moody’s, agentic AI changes the very nature of what a company can offer its clients.

“With a chatbot, you’re just having a conversation about a topic. Agents can actually perform tasks that humans would normally do,” she said.

Moody’s has begun developing agents to automate parts of its customers’ work directly, Pieretti said — for example, generating credit memos and financial analyses. Instead of retrieving one data point at a time, users can configure an agent to pull information from multiple systems, assemble the correct sections, and deliver a finished report at the click of a button.

“It shifts us from being a supplier of insights to being a workflow partner,” she said. The result: AI that doesn’t just inform decisions but helps execute them.

Building the foundations: governance and trust

At IBM, Matt Lyteson, CIO of Technology Platform Transformation, is applying the same principle on a global scale. His team is embedding agentic AI into every layer of the company’s operations — HR, IT support, procurement, and sales — serving 280,000 employees worldwide.

“Chatbots handled tasks through rigid, step-by-step flows that could easily break,” Lyteson said. “Agentic AI transforms that, enabling systems that reason through a process dynamically. That’s where the future of work is heading.”

One of IBM’s first success stories came with password resets — an unglamorous but ubiquitous use case. Two agents now collaborate: one triages the request, while the other verifies credentials and performs the reset, all under the company’s identity-and-access-management system. Each agent has its own digital identity, ensuring audit trails and preventing impersonation. “It’s a good example of multi-agent collaboration anchored in enterprise security,” Lyteson said.

Those same principles now underpin IBM’s broader Enterprise AI Platform, built on watsonx Orchestrate. The company’s AskHR, AskProcurement, AskSales, and AskIBM systems all rely on small, specialized agents operating within a unified governance framework. Every IBM employee interacts with these agents daily, likely making it one of the largest agentic AI deployments in the world.

The payoff has been dramatic. According to Lyteson, IBM’s AskIT system now resolves 82% of support requests without human intervention, freeing IT staff to focus on complex issues and allowing IBM to close its IT Service Desk phone lines. “We’re now focused on trust and collaboration — humans working confidently alongside multiple agents,” he said.

Responsible intelligence and the next phase of AI

Murali Swaminathan, CTO of IT services firm Freshworks, believes this new age of agentic AI must be guided by responsibility as much as innovation. He describes AI’s evolution in three stages: traditional chatbots, which were scripted and brittle; agent-assist systems, which indexed knowledge for humans; and now agentic AI, which understands context and acts on it. “It’s like moving from guided driving to full self-driving,” he said.

The company’s Freddy AI platform, launched in 2018, has evolved from chat support into a system that automates end-to-end workflows. In HR, for instance, an employee can request vacation time, and the agent determines which HR system to query, checks policy and balance, and executes the request. “It’s about reasoning and action, not just retrieval,” he said, adding that customers such as UK-based Frasers Group are already deflecting about a quarter of support cases using these agentic workflows.

Swaminathan emphasized that responsible AI isn’t a marketing promise — it’s a technical discipline. Freshworks’ Freddy Trust Framework embeds fairness, transparency, and privacy into every agentic workflow, he said.

The framework includes profanity and content filters, automatic masking of personally identifiable information, and rules that prevent customer data from being stored beyond an active session. Clients can also add their own guardrails. “Every deployment is designed to protect user data by default,” he said.

Freshworks has also launched the Freddy Agentic AI Studio, a no-code development environment where businesses can build and deploy agents safely. Templates, preconfigured prompts, and embedded filters make experimentation easy but controlled. “We serve everyone from small businesses to large enterprises,” Swaminathan said. “Simplicity and control must coexist.”

He calls this philosophy safe empowerment — democratizing AI while preserving trust. “Our goal,” he said, “is to help organizations adopt AI quickly and confidently — with guardrails, clarity, and simplicity at every step.”

Taking the leap from chatbots to agentic AI: A practical roadmap

Agentic AI isn’t a software upgrade — it’s a redesign of how digital work gets done. Each of the leaders interviewed for this story emphasized that success depends as much on data and governance as on culture and experimentation. Before moving beyond chatbots, IT directors should ask not only “Can we do this?” but “Where should we start — and how do we do it safely?”

Start small — and pick the right problem.

Flores at SuperWebPros recommends beginning with what he calls a “four-out-of-ten pain point” — a problem that’s mildly frustrating but not business-critical. “You want a PR win, not a huge risk,” he said. A 90-day pilot should aim to prove value quickly and visibly. Early success builds momentum and creates internal advocates.

Pieretti at Moody’s agrees: start with repeatable, well-defined workflows that deliver measurable value. “Don’t try to boil the ocean,” she advised. “Use genAI where processes are consistent and automation clearly adds business impact.”

Build on strong data and governance foundations.

IBM’s Lyteson warns against “AI sprawl” — dozens of uncoordinated pilots touching sensitive data. “Start with an enterprise AI platform that enforces identity, access, and auditability from day one,” he said. IBM’s Enterprise AI Platform gives each agent its own digital identity, mirroring employee permissions and ensuring accountability.

Swaminathan at Freshworks applies a similar principle through the Freddy Trust Framework — embedding fairness, privacy, and transparency into every agentic workflow. “With great power comes great responsibility,” he said. “Guardrails aren’t optional; they’re architectural.”

Shape culture, not just code.

Flores pointed out that human adoption is often harder than technical integration. “People resist change,” he said. “We name our agents — Marco, Betty, Harry — to make them feel like teammates rather than threats.”

Pieretti has seen the same challenge at Moody’s. “The key is shifting the mindset from ‘AI will replace me’ to ‘AI will empower me,’” she said. Training, communication, and co-creation help employees feel like part of the transformation instead of victims of it.

Adopt an iterative, governed rollout.

Both Lyteson and Swaminathan advocate continuous monitoring and versioning — agent 1.0, 1.1, 1.2 — with each release tested for drift, bias, and reliability. Pieretti’s team at Moody’s performs adversarial “jailbreaking” tests before and after deployment to ensure agents behave safely under pressure.

Swaminathan advises measuring success with hard metrics like deflection rates, resolution time, and user satisfaction. “There’s no plug-and-play AI,” he said. “Start small, measure results, and scale confidently.”

Ask the right readiness questions.

Before committing to agentic AI, IT leaders should assess four fundamentals:

Strategy: Have we identified use cases where automation will yield measurable results?

Data & Integration: Are our systems well-documented and accessible through secure APIs or metadata?

Governance: Do we have clear guardrails for identity, permissions, and audit trails?

Culture: Do we have internal champions who will model productive, responsible use?

Across all four organizations — SuperWebPros, Moody’s, IBM, and Freshworks — one message stands out: agentic AI thrives where governance meets imagination. Chatbots respond; agents reason and act. But they can only do so safely in environments built for trust, transparency, and collaboration. IT leaders who invest early in those foundations will be the ones to turn AI from a talking tool into a true digital colleague.From chatbots to colleagues: How agentic AI is redefining enterprise automation – ComputerworldRead More