What Claudeforce Signals For Every Packaged Software Vendor

Vivek Acharya is an AI strategist, author and AI ethics assessor. Advancing AI agents, process automation and responsible AI.

gettyIn late August, Salesforce and Anthropic announced Claudeforce, a partnership that places the world’s largest CRM inside Claude, Anthropic’s AI assistant. It ships with 37 prebuilt sales skills. A seller can review pipeline, prepare for meetings and update deals without ever opening Salesforce.

The product and framing are both interesting. Salesforce’s own leadership described the value of the platform as the data, the metadata and the years of encoded workflow—not the user interface. Its executives now say that when people use Salesforce through an agentic interface instead of the traditional screen, they use the platform more, not less. A company that spent nearly three decades getting the enterprise to work inside its screens is now betting that the screens were never the point.

Oracle makes the same argument for its Fusion Agentic Applications: Agents that are native to the transactional system can execute in real time under the system’s own policies, approval hierarchies and permissions. SAP’s current architecture pairs a conversational surface that generates application experiences on demand with a company-memory layer built on its process foundation, plus a hub that governs both its own agents and third-party agents.

Put these moves together, and the direction is clear: The packaged application vendors believe the durable asset is the system of record, the process semantics and the governance model. The assistant vendors believe the durable asset is the front door where work starts. Both can be right at the same time, and that is exactly what is reshaping the market.

For decades, a packaged application was a database wrapped in an opinionated interface. The interface did quiet work. It constrained users to legal actions. It enforced a shared definition of every metric it displayed. It gave two people the same artifact to argue over. When an agent becomes the user, all of that has to be delivered another way.

Distribution changes first. Agents do not sign in and click through screens. They call APIs and invoke tools. A vendor whose functionality cannot be called—through APIs and open standards such as the Model Context Protocol—is invisible to the fastest-growing category of user. Salesforce built exactly this layer months before the partnership: Its platform is now exposed to any agent as APIs, MCP servers and command-line tools.

The packaging unit changes with it. What shipped with Claudeforce was not screens but skills: packaged judgment about how to do a specific task well against the platform. Expect every vendor’s roadmap to shift from modules and screens toward skills and machine-readable definitions of what its data means.

Identity and governance become the product. When an agent inherits the acting user’s permissions, the vendor’s permission model becomes a selling point, and its weaknesses become the customer’s risk at machine speed. The vendors that win regulated industries will be the ones that can answer precisely whose authority an agent acts under, what it is allowed to write and what the audit trail shows afterward.

Pricing has to move as well. Per-seat licensing loses coherence when agents, not people, consume the software. Consumption metering is the successor, and it brings a new problem: When the application vendor and the model vendor are different companies, the customer gets two meters and two invoices for one workflow. No vendor has solved that yet. The one that does will have an advantage worth more than any feature.

Enterprise work is consolidating into a small number of AI front doors—Claude, Copilot, ChatGPT and their peers—plus the chat platforms where teams already live. Inside them, interfaces are increasingly generated on demand.

That makes the interface disposable, and it raises the value of everything the interface used to guarantee. If every view is generated fresh, the definition of every metric has to live in a governed layer outside the interface. Otherwise, 12 executives will bring 12 self-built dashboards to the same board meeting, each technically correct and none reconcilable with the others. If answers land in shared channels, someone has to decide who is allowed to see them, not just who was allowed to retrieve them.

In my work, the question I hear has changed. It is no longer which suite has the better interface. It is which layer reconciles them. And because no vendor can govern its competitors’ systems, the layer that reconciles definitions, authority and cost across an estate will not come from any of them. Clients have to build and own it.

My read is that this plays out in three moves over the next couple of years.

First, the interface layer consolidates. Daily work settles into a small number of assistant front doors, and the application screen becomes what the admin console is today—a place specialists visit, not a place work happens. Competition among vendors shifts from how their screens look to how well their platforms behave when called: the precision of their permissions, the quality of their skills, the completeness of their audit trails.

Second, the commercial model follows the usage. Seats give way to consumption, and the two-invoice problem gets solved—either by application vendors bundling model inference into a single bill or by a new layer that gives finance one view of what agentic work costs. Whoever provides that view first will have a seat at every renewal.

Third, a neutral layer emerges above the suites, and I think this is the least appreciated part. Every vendor can govern its own agents. None can govern its competitors’. The definitions, the authorization policy and the cost visibility that span an estate will end up owned by the enterprise itself, outside every suite.

Where the value sits? It is not in the screens. The next phase of this market belongs to whoever makes the system of record, its processes and its governance safely available wherever work happens—and to the enterprises that build the layer that reconciles them all.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?