What AI Agent Adoption Challenges Mean for the Enterprise
AI agent adoption challenges in the enterprise are the practical and organizational barriers that prevent intelligent agents from moving beyond isolated experiments into secure, trusted, high‑scale use across real business operations. For IT and digital leaders, these challenges are no longer abstract. Salesforce Futures VP Mick Costigan says customers are trying to “turn fast-moving capability into trusted, organization-wide value” while coping with uncertainty around models, data, regulation, sovereignty, cost and the future of work. At the same time, Salesforce is pushing its vision of the “agentic enterprise,” where agents can reason, act, access tools and automate or support workflows across functions. Understanding why that vision is hard to implement matters more than reading about the latest model benchmark. The gap between what models can do and what enterprises can safely deploy is now a core strategic question for every AI roadmap.
Challenge 1: Predicting Agent Capabilities in a Fast-Moving Landscape
The first barrier is uncertainty about how capable AI agents will become and how quickly. Costigan describes model progress as feeling like “the beginning of an exponential rate,” yet CIOs cannot see clearly which frontier models will matter most, or for how long. This makes long-term bets on a specific stack risky and slows enterprise AI deployment. Agents are emerging as the “killer app,” but their usefulness depends on more than raw model performance. Leaders must weigh whether to commit to one ecosystem or design for a future where multiple models power different types of agents. Costigan advises customers to run on two tracks: deliver near‑term ROI from practical use cases while exploring how more powerful agents might reshape business structures and processes. This dual approach helps organizations avoid both paralysis and reckless over‑commitment.
Challenge 2: Connecting Agents to Enterprise Data, Tools and Guardrails
Even if models improve rapidly, the second barrier is bringing agents into complex enterprise environments. This means connecting them to systems of record, tools, permissions, governance frameworks and user interfaces without breaking security or compliance. Costigan points out that “getting data right, getting access to tools right, getting the interface right” is far harder than spinning up a model. Salesforce’s answer is the Agentforce “agentic harness,” a layer around large language models that adds context access, policy enforcement and integration with business applications. The analogy is an early car: the engine alone is not enough; you also need brakes, steering, dashboards and rules of the road. Enterprise agents must work with zero data retention when required, pull from trusted data rather than hallucinate, and leave auditable traces of their actions. Without that harness, pilots will stall before reaching organization‑wide deployment.
Challenge 3: Redefining Human Roles and Operating Models
The third barrier is human: what people do when agents can reason, decide and act. Enterprises cannot scale agentic systems without rethinking jobs, skills and accountability. Costigan’s third core question is simple but disruptive: “what do humans do?” As agents automate more steps in sales, service, marketing and commerce, organizations must redesign workflows so humans supervise, handle exceptions and focus on higher‑value work, rather than fight against automation. This is evident in Salesforce Agentforce Commerce, where Buyer and Merchant Agents handle tasks like conversational ordering or catalog organization in plain language. The human role shifts from manual processing to oversight and strategy. Clear ownership, new performance metrics and training in working alongside agents become mandatory. Without an explicit plan for the future of work, resistance from teams and unclear responsibilities can block AI agent adoption, even when the technology is ready.
Agentforce Commerce: Real-World Testbed for Agentic Enterprise Barriers
Salesforce’s latest Agentforce Commerce updates show how these barriers surface in live B2B and B2C deployments. The Buyer Agent conducts procurement via WhatsApp and SMS, translating messages like “Need 40 cases of the 16‑oz fasteners, same as the March order” into product confirmation, contract pricing and a completed order without portals or phone calls. On the B2C side, a Shopper Agent and agentic commerce search aim to personalize discovery, backed by integrations with ChatGPT, Gemini and Google AI. According to Digital Commerce 360, 78 of the Top 2000 online retailers use Salesforce ecommerce technology, accounting for more than $192.60 billion in web sales in 2025. That scale sharpens requirements for reliable data access, clear merchant‑of‑record status and unified order management. As Agentforce rolls out, customer feedback will continue to expose the organizational, technical and operational hurdles that define real‑world agentic enterprise barriers.
