AI was supposed to be the headless horseman that signified the apocalypse of software as a service companies, but Salesforce has joined forces with the enemy and is now allowing your agents to access its core technology, minus the user interface. Plus its own agents, of course: probably preferably, to Salesforce. But you can now bring your own agents to the party. For me, that would include both OpenClaw and NanoClaw .

Today Salesforce launched AgentForce Operations. The key idea: agents can do the busywork for you … deep inside the messy plumbing of the enterprise: supply chain, procurement, finance, claims, underwriting, IT provisioning. The promise: agentic workflows can cut cycle times by 50-70% and manual data entry by 80%.

The big news: you can bring your own agent.

“Customers can integrate their own agents if they’re approved internally,” Salesforce SVP of product Sanjna Parulekar told me via email. “Agentforce Operations comes with built-in task-specific agents, but it’s designed to be flexible: it uses an LLM-based planning layer to interpret instructions and a deterministic execution engine to carry them out across tools and systems. If a customer has external agents, they can be invoked via API and seamlessly plugged into a workflow, with results passed back into the orchestration layer.”

Like many of you, I’m embracing agents. A NanoClaw agent on a Mac mini recently helped me come back from email bankruptcy by deleting 29,224 emails older than 30 days and answering dozens more for me. And an OpenClaw agent in the cloud is building an ecosystem mapping for me of an industry I’m studying. The massive benefit: extended capability, but also actioning items that otherwise just never get done.

Salesforce, of course, has its own agents, and you can use them easily to push digital paper, complete workflows, save information, route tasks, verify details and much more.

The technical bones come from Regrello, the supply chain workflow company Salesforce acquired and has spent the last year integrating. Regrello’s core insight was that workflow automation tools have always broken at system boundaries: they route tasks between humans and applications, but they don't actually finish the work.

Agentforce Operations is the generalization of that idea across every (or most) back-office process.

The mechanism is something Salesforce calls a “Blueprint,” a structured, end-to-end workflow with defined start and end states, versioning, and variants for different regions or product lines. Inside a Blueprint, work flows through specialized agents that handle specific jobs: extracting data from tax returns, checking inventory, validating compliance rules, chasing missing signatures.

This is a good idea: give an agent too many jobs with too wide a variance and you’ll likely get drift in quality over time. Focusing – and giving agents tight guardrails – help constrain an inherently probabilistic technology to deliver mostly deterministic results.

To that end, Salesforce ships more than 30 out-of-the-box Blueprints for jobs like invoice auditing, onboarding, and PO rescheduling. Customers can also generate new ones from existing process documentation.

Oh: and agents leave breadcrumbs: every action an agent takes is mapped back to the Blueprint, creating what Salesforce calls "radical transparency" and what a compliance officer might call a requirement.

Importantly, you can chain Blueprints together, which means you can create complex agentic workflows from discrete, bounded, and generally reliable components. Parulekar confirmed that customers can string agentic workflows and also combine human and agentic tasks inside a single deterministic blueprint. This is a pretty significant bridge from plain vanilla "an agent does a thing" to a much bigger wow: actual end-to-end process automation with intelligence and – where needed – a human in the loop.

If you’re pretty used to burning tokens in agentic workflows and AI builds, there’s going to be something new to learn: Parulekar says pricing is based on the exact Blueprints you use. Since those are structured, end-to-end workflows with a defined beginning and end, she says, this model “offers extensive flexibility through multiple variants tailored to specific scenarios - such as regions or product lines - and supports unlimited versioning to facilitate continuous process improvement.”

A likely translation of that is that you’ll pay some fraction of what your agentic process automation saves you.

Of course, Salesforce isn’t the only company with this idea. ServiceNow has been arguing essentially the same thesis for the last year with a pitch that it’s becoming the "AI control tower" for the agentic enterprise. CEO Bill McDermott has been blunt about the strategy, saying “there’s going to be many millions more of agents than there are human beings.”

Like every massive change, some impacts are unexpected.

"When companies design processes to be human-first, they think about how to remove steps to improve and speed up processes, but when companies design their processes for AI, it opens up new possibilities," Parulekar told me. "We've seen customers ADD steps to their processes to increase accuracy, compliance and improve employee and customer experience, while still drastically reducing cycle times. Agentforce Operations is giving customers the flexibility to build what is best for their business without the constraints of headcount, working hours, or turnaround times."

That might be a more important finding than the cycle-time numbers. Twenty years of process reengineering orthodoxy has assumed that fewer steps equals better processes, because every step costs human time. When agents are doing the work, that constraint inverts. You can add a verification step, a compliance check, a customer notification — things you’d never staff a human to do — and still finish faster than before. The end result might be better quality, better compliance, better profitability … without extra delay.

This might even be a counter-argument to the almost-ubiquitous "AI replaces humans" narrative given the recent spate of layoffs. An alternative framing might be that AI removes the budget ceiling on process quality.

Time, as they say, will tell.