January 05, 2026
2025 Was About Agentic AI. What Comes Next?
2025 was the year artificial intelligence moved deeper into doing. What began as a promise of autonomous agents in research papers and visionary roadmaps became a tangible force in customer experience, enterprise automation, and strategic toolkits across industries.
“Agentic AI” describes AI systems capable of autonomous reasoning, understanding a goal, deciding how to achieve it, and acting without constant human direction. In CX, this changes AI from an assistant that supports agents to a system that can actively move cases toward resolution.
By early 2025, key building blocks were already in place: reasoning-focused model architectures, growing industry attention on autonomy, and early attempts to enable AI systems to carry tasks through to completion. The following timeline traces how those pieces came together over the year, from groundwork to moments that forced enterprises to take agentic AI seriously.
January: Emergence and Technical Framing
The first widely available agent-style capability appeared with the launch of Operator, a feature in ChatGPT that could execute real tasks, such as filling forms and scheduling appointments, through autonomous browser interactions. This marked one of the earliest instances of agentic behaviour being embedded in a mainstream product, even if the term “agentic AI” wasn’t yet ubiquitous in headlines. Industry commentary also began framing 2025 around AI agents and autonomous problem-solving, setting expectations for the year ahead.
At the World Economic Forum in Davos, SAP’s CEO publicly declared agentic AI a meaningful evolution in 2025 and said the company planned to launch agentic systems in sales and supply chain domains, moving the conversation from theory to business application.
In retrospect, January was the month when the lexicon of autonomy entered mainstream consciousness, and when CX and enterprise leaders began internal discussions about moving beyond assisted workflows.
February: Concept Meets Reality
In February, Salesforce deepened its partnership with Google Cloud to bring Gemini models into Agentforce, giving its platform stronger reasoning and multimodal capabilities. Before that, the CRM giant previewed autonomous agents in its Agentforce 2.0 release in December 2024.
The Google Cloud partnership mattered less for the features themselves than for what it meant for the industry. Major CX platforms were preparing AI agents to take on more complex, end-to-end work inside contact centres, rather than limiting them to assistive roles.
March: Agentic AI Breaks into the Open
At Zendesk’s Relate conference, the company unveiled its Resolution Platform, revealing AI agents as systems that could reason across context, trigger actions, and resolve issues end-to-end rather than simply assist human agents. The emphasis was on outcomes: closing tickets, coordinating steps across systems, and reducing customer effort.
The same month, Cisco brought Webex AI Agent into general availability, presenting it as an autonomous layer capable of handling customer interactions, taking action across workflows, and escalating only when needed. At Enterprise Connect, Zoom also introduced specialised AI agents for Zoom Contact Center, focused on task completion and orchestration rather than scripted automation.
Across these announcements, the common thread was intent. Platforms were openly presenting AI agents as operational components of contact centres, expected to decide, act, and resolve, with humans stepping in only when autonomy reached its limits.
April: Broader Adoption and Strategic Integrations
Agentic AI slowly became part of actual go-to-market initiatives that pointed toward enterprise adoption. Capgemini and Google Cloud announced a strategic effort to bring agent-style AI into customer experience operations at scale, combining Google’s advanced reasoning models with Capgemini’s systems integration capabilities to help clients deploy autonomous workflows across service and support functions.
April also saw Teleperformance unveil partnerships with AI vendors to accelerate the adoption of autonomous agent frameworks in service delivery, signalling that large outsourcers were preparing their operations for agentic AI at scale.
May: Practical CX Plans
May brought the first enterprise-oriented agentic AI initiatives beyond conceptual framing, with vendors outlining how autonomous capabilities would play out in customer experience technology.
One of the most tangible examples came from Dialpad, which in mid-May teased a broader agentic AI platform aimed at pre-emptive customer service, moving beyond reactive bots toward systems designed to anticipate needs and take autonomous action inside contact centres.
Around the same time, TaskUs expanded its agentic AI work through partnerships with Decagon and Regal, extending its agentic AI consulting offering into autonomous customer support delivery. The focus was on deploying systems that can complete end-to-end tasks across digital and voice channels, with human oversight built in to manage operational risk.
Also, Microsoft used Build 2025 to articulate a bigger industry push toward AI agents, emphasising open standards and composable agentic workflows as part of an “agentic web.”
June: Scepticism and Realism in Enterprise Reporting
June brought a dose of realism to the agentic AI conversation. Gartner warned that a large share of early agentic AI initiatives were at risk of being abandoned due to high costs, unclear ownership, and weak links to measurable business outcomes. Autonomy alone was not a strategy, and many projects were moving faster than organisations could govern them.
Meanwhile, leading CX platforms formalised how agentic AI would actually be deployed. Genesys introduced Genesys Cloud AI Studio, a controlled environment for building and managing AI agents with explicit guardrails around data use, escalation, and oversight. NICE and Five9 also advanced agentic capabilities during the month, placing governance, trust, and operational accountability at the centre of their CX roadmaps rather than treating autonomy as a standalone feature.
September: Enterprise Pilots and Practical Platform Advances
By September, agentic AI had moved into both pilot deployments and expanded platform tooling. Citigroup continued internal pilots of AI agents designed to synthesise research and manage multi-step workflows, illustrating agentic systems’ expanding scope in knowledge work.
On the vendor side, Zoom used its Zoomtopia 2025 conference to unveil AI Companion 3.0, enhancing its AI agent stack with deeper context-aware reasoning and retrieval across meetings, calls, and shared documents.
Together, these developments showed how agentic AI was being used in practice and how platform capabilities were catching up to enterprise expectations. The question was no longer whether agentic AI could work, but how far organisations were prepared to let it operate on its own.
End of Year: Strategic M&A and Big Tech Bets
By the end of 2025, agentic AI had moved from early experiments to major strategic commitments by the largest tech companies. In December, Meta agreed to acquire Singapore-based AI agent startup Manus for more than $2 billion, with the deal finalised at the time of writing. The acquisition brings a general-purpose autonomous agent into Meta’s product ecosystem and underlining how serious big tech is about agents that can act.
The same period saw agentic behaviour surface more directly in end-user platforms. Microsoft continued expanding agent-style capabilities inside Windows 11, where Copilot was increasingly praised as an operating-system-level assistant able to act across applications, settings, and workflows rather than remain a passive helper.
In parallel, CX vendors such as Dialpad moved closer to using agentic AI in live customer support discussed earlier in the year, reinforcing that autonomous behaviour was something customers would actually use, not just hear about.
Closing Thoughts
2025 left little ambiguity about what agentic AI is and isn’t. Over the course of the year, it moved out of theory and into real operational settings, particularly in customer experience, where autonomy is tested quickly and publicly.
The year also exposed the limits of autonomy. Agentic AI handled structured tasks well and reduced manual effort across workflows, but it struggled in areas that require judgment, accountability, and trust. These limits pushed organisations to deal with very practical issues, such as who is responsible when an AI makes a mistake, how much freedom it should have, and when a human needs to step in, all of which show up fastest in customer experience.
