Salesforce Just Made the “Own Your AI Layer” Argument a Lot Harder to Ignore

We have been having the same conversation with nearly every client over the past few months. It goes something like this:

“Who owns your AI orchestration layer? Do your own agents orchestrate across your systems of record, or does each vendor own that layer and bill you separately for the privilege?”

Many CIOs and procurement leaders are considering a departure from the norm: own it yourself. One AI bill instead of many. No lock-in to a single vendor’s agent stack. Full visibility into what your agents are actually doing and what it costs. There are many unknowns regarding the work involved to execute this approach, but there is definitive interest (if not momentum) behind the notion.

This week, Salesforce made that path significantly more credible. They’ve also raised several questions that will test the viability of owning your own orchestration layer.

What Salesforce Actually Announced

At its TDX 2026 developer conference in San Francisco, Salesforce launched Headless 360.

The short version: they have exposed their entire platform as APIs, MCP tools, and CLI commands so that AI agents can operate Salesforce completely without a browser or a human login.

Your Claude agents. Your ChatGPT agents. Your own orchestration layer. They can now call directly into 25 years of Salesforce data, workflows, and business logic without Agentforce as the middleman.

The release ships over 100 new tools immediately, including more than 60 MCP tools and 30 preconfigured coding skills. Salesforce also released Agent Script, an open-source language for defining agent behavior deterministically, which signals that they are serious about making this infrastructure-grade rather than demo-grade.

This is arguably the most important move a large enterprise SaaS provider has made in the AI era, and an indication of what’s to come. Salesforce is doing more than simply building agents on top of their platform and selling you access to those agents. They are making the platform itself available to whatever agent runtime you choose to run.

Why This Matters for the “Own Your Layer” Debate

The core tension in enterprise AI right now is this: every major SaaS vendor is simultaneously building agents and positioning their agent platform as the coordination layer across the enterprise. If you let that happen, you end up with five different agent layers, five different pricing models, and no unified view of what any of it costs.

The alternative, which is gaining real consideration in some companies, is building or designating your own orchestration layer and having your agents call into vendor systems via APIs rather than running inside each vendor’s walled garden.

Until recently, Salesforce was a hard platform to adopt this approach with. The depth of workflow logic, process automation, and customer data inside most Salesforce implementations made it difficult to run meaningful agent workflows against the platform without going through Agentforce.

Headless 360 could change that. It is not a workaround. It is Salesforce explicitly opening the infrastructure to external agents.

The Catch Nobody Is Talking About: Consumption Pricing

Sound too good to be true? Possibly. As companies debate the feasibility of “own your own orchestration layer,” they need insight into two things:

1. At what cost?
2. How do we predict and manage these costs?

Salesforce has not announced how Headless 360 will be priced. But we can expect premium, consumption-driven pricing once SKUs appear. It will likely be tied to Agentforce and Data 360 usage rather than traditional seat licenses as Salesforce finds a way to meter and charge for access.

Salesforce has already shifted its Agentforce pricing from per-seat licenses to consumption-based billing. That shift reflects an AI reality: when agents are doing the work, seats are the wrong unit. But consumption pricing pushes the visibility problem entirely onto the buyer.

When humans consumed software, you forecasted seats. When agents consume it, you are forecasting token burn rates across workflows that do not behave the same way twice. (at least for right now – the repeatability factor is improving with every iteration).

Now imagine five platforms with consumption models and your own agents calling into all of them autonomously. If you do not own the intelligence infrastructure to govern that spend, you have traded vendor lock-in for vendor chaos.

When we sit down with clients on this question, we talk through two paths:

• Path A: Rely on each vendor’s agent layer. Each platform (Salesforce, ServiceNow, Workday, Microsoft, etc.) orchestrates within its own domain. You get faster time-to-value within each system but no unified orchestration, no consolidated cost visibility, and a bill from each vendor’s agent platform on top of your existing license spend.

• Path B: Own your orchestration layer. You run a central agent runtime, you call into each platform via APIs, and you maintain visibility and control over what agents are doing and what it costs. Much like the traditional buy versus build debate, there is higher upfront investment in architecture (although this cost is becoming cheaper for teams that know how to effectively wield AI coding tools). The byproducts? A significantly better position at renewal time and in managing total AI spend.

The pattern we are seeing is that organizations considering Path B are doing it not because they have a fully formed strategy, but because the math on Path A is getting difficult to defend to the CFO.

What NPI Is Advising Right Now

Headless 360 is fresh to market and there is still a lot to be evaluated (pricing being an obvious example). But it’s worth a serious look for any organization where Salesforce is a core system of record and where you are already building or planning to build agents.

Before you move, get clear on three things:

1. Your consumption baseline. What does your current Salesforce spend look like in terms of API consumption, automation runs, and Agentforce credits? If you do not have that number, you are not ready to let agents run autonomously against the platform.
2. Your contract terms. Consumption pricing with credit multiplier provisions is not the same as fixed-rate API access. Know exactly what triggers cost increases and where your protections are.
3. Your governance layer. Owning your AI orchestration layer will only be an advantage if you can see what it is doing. That means logging, cost attribution, anomaly detection, and workflow-level spend visibility across every system your agents touch.

Salesforce has made the platform available. Whether you can take advantage of it in a way that is actually sustainable comes down to whether your own house is in order first.

NPI helps enterprise technology buyers optimize AI spend. Interested in a conversation? Contact us.