OpenAI Frontier arrives with the language of enterprise agents, but its larger target is the architecture—and pricing logic—of business software.
The platform sits above an organisation’s data warehouses, CRM systems, ticketing tools and internal applications. It provides shared business context so agents can be given identities, permissions and performance reviews much like digital coworkers. Frontier can manage agents from OpenAI, internal development teams and third parties.
That design challenges a software market built around humans logging into separate products. If an agent completes work across several systems, the value of each individual seat becomes harder to defend.
One context layer instead of another silo
Agents deployed one by one can add exactly the fragmentation they were meant to remove. Each needs its own integrations, definitions and governance. Different agents may develop conflicting views of customers, products or approval rules.
Frontier’s answer is a semantic layer that supplies a shared understanding of how the organisation works. Agents can reference the same business context without every team reconstructing it independently. OpenAI Applications CEO Fidji Simo described the frustration of integrating tools at Instacart only to reinforce silos because they could not communicate.
Early results cited by OpenAI are substantial. An investment firm reportedly removed more than 90% of the administrative time from parts of its sales process. A technology customer saved 1,500 hours per month in product development, while a manufacturer compressed production optimisation from six weeks to one day.
These claims still need to be interpreted carefully. Time saved in one step is valuable only if quality, exceptions and downstream work remain controlled. Frontier’s ability to assign permissions and review performance is therefore as important as its ability to connect systems.
Agents destabilise the software seat
Traditional software-as-a-service revenue maps access to headcount. A company pays for each employee who uses a CRM, workflow platform or productivity application. An agent that works across those tools breaks the connection between one user and one licence.
The market has noticed. Salesforce shares declined more than 27% during the year despite strong quarterly revenue and rapid Agentforce sales, as investors worried about agentic disruption and reacted to weaker-than-expected guidance.
Incumbents are adapting pricing. Salesforce introduced an all-you-can-use Agentic Enterprise License Agreement. ServiceNow has moved some agent products toward consumption pricing, while Microsoft combines usage-based pricing with per-user Copilot Studio plans.
This is more than a billing adjustment. Software that becomes an invisible system of record may remain essential while losing the visible workflow that justified a premium seat. Vendors must prove that their data, controls and embedded agents deserve a central role even when another platform orchestrates the work.
The intelligence layer is now contested territory
Salesforce and ServiceNow argue that agents belong close to the systems holding the data. Their existing security, compliance and customer relationships can provide faster deployment and institutional trust. Salesforce calls Agentforce an operating system for the agentic enterprise, while ServiceNow’s AI Control Tower governs agents from different sources.
OpenAI takes the overlay position. Frontier connects existing systems through open standards so companies can add agents without replacing the infrastructure underneath. The benefit is reach across vendors; the risk is placing another powerful layer between an enterprise and its operational applications.
Most large organisations already use Salesforce, ServiceNow and Microsoft together. Frontier could become the connective tissue among them, or it could gradually absorb the workflows through which users experience their value.
Frontier remains limited to selected customers, with broader access expected later and no public pricing. OpenAI also enters the contest without the decades of contracts and trust held by established enterprise vendors. It brings stronger identification with frontier-model capability and an argument that intelligence should span the whole company.
The immediate winner may not displace SaaS systems at all. It may force them to become more open, measurable and economically aligned with machine work. Once agents—not employees—operate the interface, enterprise software must explain what customers are paying for when the seat is no longer the centre.
