From chatbots to agentic enterprises: the real AI challenge
Enterprise AI agents are software systems that can reason, take actions, access tools, and automate or support complex business workflows at scale, moving far beyond simple chatbots to become embedded digital workers inside customer service, sales, operations, and other functions across an organization. That sounds like the future, but the headache is already here. As models race ahead, enterprise leaders face a harsher problem: turning fast-moving capability into trusted, organization‑wide value without being buried under doubts about data, regulation, sovereignty, cost and the future of work. Big tech firms are pouring billions into agentic AI, chasing a market projected to grow from USD 10.9 billion (approx. RM50.1 billion) in 2026 to USD 182.9 billion (approx. RM841.3 billion) by 2033. The money is chasing a clear thesis: agents will reshape how business gets done, but only if teams can overcome three core challenges.
Capability vs reality: why smarter agents do not equal smarter enterprises
Mick Costigan, VP of Salesforce Futures, says customers keep coming back to three questions: how good agent capabilities will become, how to bring them into the organization, and what humans will do once they arrive. These questions expose a widening gap between model capability and AI agents in use. Models may feel like they are at the start of an exponential curve, but enterprises move slower because they depend on reliable data, access to tools, sensible interfaces and measurable productivity gains across the organization, not just for a single user. In practice, enterprise AI agents must operate in dense environments: systems of record, permissions, governance policies and legacy infrastructure. Frontier models alone cannot cross that terrain. The harsh truth is that many AI agent deployment efforts stall not because the engine is weak, but because the organization lacks a safe way to let it run.
Integration and the agentic harness: the new systems engineering
Agentic AI implementation is now less about picking a frontier model and more about building what Salesforce calls the "agentic harness" – the layer that connects agents to data, tools and workflows with guardrails and context. Costigan compares early large language models to an 1886 motor wagon: an engine bolted to three bicycle wheels and a garden bench. Enterprise value came later, when cars gained brakes, steering, dashboards and safety systems. The same story is playing out with enterprise AI agents. To move beyond chatbots toward agents that can qualify leads, manage bookings and process transactions inside channels such as messaging apps, companies must wire them into existing sales, booking and customer management systems. Cloud giants are embedding agents into resource planning and customer relationship platforms, while other players push multi‑agent frameworks that span departments. The winners will be those who treat integration as systems engineering, not a collection of demo scripts.
Governance and measurement: power with accountability
As agents gain autonomy, governance and performance measurement move from side issues to existential ones. Enterprise AI agents must run on trusted context, with access to the right data and guardrails that keep outputs auditable. That means zero data retention requirements for frontier models, clear permissions, and governance layers that can withstand regulatory scrutiny. Convenience comes with trade‑offs: the more useful an agent becomes, the more influence it has over the interaction and, in time, the customer relationship. Customers may enjoy instant answers when they message a store, receive stock checks and alternatives within seconds, and small businesses may finally offer 24/7 support without extra staff. But enterprises that fail to instrument these systems – tracking accuracy, bias, escalation rates and downstream outcomes – risk handing critical decisions to opaque automation. Governance cannot be an afterthought; it must be built into the harness as tightly as any API.
Redesigning work: agents as colleagues, not replacements
The toughest challenge is not technical but human: what humans do in the agentic enterprise. As agents shift from conversational assistants to systems that reason, act and automate across business functions, teams must rethink roles, skills, workflows and accountability. Meta’s move into transactional agents shows the direction of travel: automation is reaching the moment where attention converts into a booking, purchase or solution to a problem, not only the pre‑sale interaction. Mark Zuckerberg’s company, long built around advertising revenue of USD 200 billion (approx. RM921.3 billion), is now seeking a place inside the customer relationship itself. This is the pattern every enterprise will face. The smart response is not to resist agentic AI, but to design for co‑work: agents doing routine orchestration, humans owning judgment, exception handling and relationship‑building. Without intentional work design, the agentic enterprise will not be trustworthy – or productive.
