Salesforce is selling Agentforce as the answer to the SaaSpocalypse, but some of its flagship use cases are still waiting for ordinary users. A University of Chicago Medicine example shows the promise of AI agents, and the operational drag that can slow them down.
Salesforce wants investors and customers to believe Agentforce is already changing enterprise software. The harder truth is more useful: in regulated industries, the path from a polished AI demo to a live customer workflow is still full of friction.
According to a Bloomberg report published on May 21, University of Chicago Medicine has not yet made several promoted Agentforce capabilities broadly available to patients, including AI-assisted prescription refills, appointment scheduling, and parking help. The delays were tied to product glitches and the difficulty of securing internal compliance approvals at the healthcare network. Patients calling the system are still met with keypad menus and human schedulers, while the chatbot remains in testing and is not visible to most web visitors.
This is not a small embarrassment around one customer video. It touches the central question facing Salesforce. The company reported fiscal 2026 revenue of $41.5 billion, up 10 percent, after growing 25 percent in fiscal 2022. Its next act depends heavily on AI. Salesforce said Agentforce reached $800 million in annual recurring revenue and more than 29,000 deals since launch, but buyers now have to separate signed demand from deployed value.
The Demo Versus The Deployment
The patient engagement use case is exactly the kind of workflow that makes AI agents sound powerful. In a controlled setting, an agent can pull a simulated record, check appointment availability, confirm refill status, and point someone toward the right parking deck. In a real hospital system, the agent has to work across electronic health records, scheduling systems, identity checks, insurance requirements, clinical permissions, and audit trails.
That difference matters. Healthcare systems do not buy software the same way a retailer adds a new support widget. A broken answer can become a patient safety issue, a privacy problem, or a compliance review. The technology may be impressive, but the organization still has to decide who is accountable when the agent acts.
Madhav Thattai, Salesforce executive vice president of Agentforce, told Bloomberg that customers often start with simpler use cases before moving into more complex automation. That is a reasonable rollout path. It is also a reminder that enterprise AI is being marketed faster than many enterprise customers can absorb it.
The Startup Wedge
For startups, this is where the opening appears. AI-native companies do not need to rebuild Salesforce. They can focus on one painful workflow, such as prior authorization, benefits verification, appointment routing, or post-discharge follow-up, then prove that narrow system works inside the messy reality of a hospital.
That focus is valuable because enterprise buyers are not only shopping for models. They are shopping for proof. A startup that can deploy a working tool in weeks, integrate with the systems already in place, and show measurable savings has a stronger pitch than a broad platform still moving through committees.
Salesforce still has a formidable advantage. It owns the customer relationship, the data layer, and the procurement channel inside many large organizations. It can put Agentforce in front of buyers that smaller vendors would struggle to reach. But distribution only carries the product so far. If users do not see reliable outcomes, the sales motion becomes a source of pressure rather than protection.
There are examples where Agentforce appears better suited to the task. SharkNinja has used automated product troubleshooting to reduce service phone calls by 20 percent this year, according to the same Bloomberg report. Kyle, Texas, also expanded its Salesforce spending as it used Agentforce to help residents report issues such as potholes and graffiti. Those are valuable workflows, but they are different from healthcare. A missed pothole ticket is frustrating. A failed refill request is more serious.
What Enterprise Buyers Should Watch
The lesson for enterprise leaders is not to dismiss agents. It is to demand a clearer line between demo, pilot, and production. A working agent should be judged by completion rates, escalation quality, compliance review time, and what happens when the system is uncertain. The bar is higher than it was for chatbots because the product is no longer just answering questions. It is taking action.
For Salesforce, the next milestone is whether University of Chicago Medicine can bring these patient-facing features into broader use. If the system rolls out AI-driven scheduling and refill support over the next year, it strengthens the company’s argument that complex deployments simply take time. If the scope narrows or the timeline keeps slipping, it gives rivals a sharper message.
For founders, the takeaway is practical. Do not wait for the incumbent to fail. Build the specific thing a large platform struggles to deliver because it is trying to serve everyone at once. The market for agents will not be won by the best demo. It will be won by the product that survives contact with real operations.
Also read: Arafura Rare Earths plans 250 million share sale backed by Australia's richest person http:// reut.rs/3RfFUVY • Reuters alleges Iran routed billions through Binance to skirt sanctions • JPMorgan moves to shed roughly 4bn in private equity-linked loans as banks curb leveraged finance risk
