Salesforce made the platform agent first on 15 April 2026. Production pricing is not public yet, but the meter is live. Read the pillar before you wire agents to production data.
Salesforce Headless 360 lets agents and code consume the platform through APIs and MCP tools instead of the browser. Salesforce announced it on 15 April 2026 and has not yet posted production pricing. This pillar covers the architecture, the meters, and the buyer side levers to pin before you build.
Salesforce spent 2025 moving its whole platform toward agents. Headless 360 is the next step. It reframes the platform as something an agent or a block of code calls, not something a person clicks.
That shift changes the licensing question. The old question was how many seats. The new question is how much agent traffic, metered in credits, and at what unit rate. This pillar answers both.
Headless 360 is an agent first way to consume Salesforce through APIs, MCP tools, and agents rather than the browser. Salesforce introduced it at TrailblazerDX on 15 April 2026, positioning the platform as a system agents drive.
Headless means the user interface is optional. An external agent, a model, or a service can read and write Salesforce data directly. The billing follows the traffic, not the login.
This is the core licensing change. When people click, you count seats. When agents call, you count consumption. Headless 360 makes consumption the primary meter for a growing share of platform use.
The change is not cosmetic. A headless design can serve customers with almost no interactive logins, which means a traditional seat forecast will badly understate the true cost. The spend simply moves from the seat line to the credit line, and it often grows on the way.
Salesforce shipped more than 60 MCP tools and more than 30 coding skills with Headless 360. The Model Context Protocol lets an external agent discover and call Salesforce capabilities as named tools.
Each tool call is a unit of work. That unit is where metering lives. A buyer needs to know which meter a tool call hits and at what rate before wiring agents to production data.
Agentforce Vibes 2.0 pairs the build experience with model choice, including Claude Sonnet and GPT 5. The Salesforce developer stack adds an Experience Layer and a DevOps Center MCP that Salesforce says can cut build cycles by up to 40 percent.
Headless 360 bills across four layers, and each layer has its own meter. Understanding which layer a workload touches is the first step to a defensible forecast.
Data 360, the platform formerly known as Data Cloud, meters on data ingested, stored, and processed. Agent traffic that resolves customer context runs through this layer, so headless agents can drive Data 360 consumption up sharply.
Watch the query and processing meters most closely. Every time an agent looks up a customer record, segments an audience, or grounds a response in your data, it touches Data 360. At agent scale those lookups compound, and the Data 360 bill can rival the Agentforce bill.
Customer 360 is the licensed application estate, still counted largely in seats and editions. Headless access reduces seat pressure for some roles but does not remove the underlying edition entitlement that governs objects and limits.
Agentforce is the agent runtime, and it meters in Flex Credits. Every agent action, every Voice interaction, and much of the reasoning draws credits. This is the layer most exposed to headless growth.
Slack is increasingly the human surface for agents. Actions triggered from Slack still resolve through Agentforce and Data 360, so Slack does not escape the consumption meters underneath it.
The four layers and how each meters
| Layer | What it is | Primary meter | Headless exposure |
|---|---|---|---|
| Data 360 | Unified data platform | Data ingested, stored, processed | High |
| Customer 360 | Licensed applications | Seats and editions | Medium |
| Agentforce | Agent runtime | Flex Credits per action | Very high |
| Slack | Agent and human surface | Seats plus underlying meters | Medium |
Headless 360 has no standalone production price list as of July 2026, so the real cost flows through Flex Credits and per user add ons. The headline is missing, but the meter is running.
The Agentforce pricing page and current order forms give the anchors buyers can rely on today.
The per user add on is the seat side of the model, near 125 dollars per user per month, or 150 dollars for regulated industries such as financial services and healthcare. It covers the human who supervises or builds agents, not the agent traffic itself.
Do not confuse the add on with the consumption meter. A team can hold a modest number of add on seats while the agents those seats supervise burn a far larger sum in Flex Credits. The two costs move independently, and both belong in the forecast.
Four line items are usually blank in the first order form, and each is a place where cost can escape. Pin every one as a named unit with a capped rate.
The standard account team framing is that Headless 360 is included with the platform and that agent capability comes with your existing entitlement, so cost is not a concern at this stage. We disagree. In roughly seven out of ten Agentforce estates we have modeled, the included framing hid a live consumption meter that surfaced only at the first true up, and by then the agents were in production and the leverage was gone. The buyer side move is to treat every agent action as a metered unit from day one, model the credit burn before you build, and lock a capped unit rate and a rate lock term in the order form. That is not how the platform is sold, and it is exactly why buyers overpay.
Source: Redress Compliance advisory engagement file, 2025 to 2026.
Headless 360 is not a free upgrade. It is a shift from counting seats to counting agent actions, and the meter starts the day the first agent goes live.
Flex Credits are the currency for agent work, and headless traffic draws them fast because one request can trigger many actions. The credit, not the seat, is the unit to forecast.
A Flex Credit is worth about half a cent at 500 dollars per 100,000. An action at 20 credits costs near 10 cents. Simple math, but the volume is what surprises teams.
Take an agent that handles 5,000 requests a day, and assume each request triggers 4 actions. That is 20,000 actions a day, or 400,000 credits, near 2,000 dollars a day before discount.
Illustrative daily Flex Credit burn
| Requests per day | Actions per request | Credits per day | Approx dollars per day |
|---|---|---|---|
| 1,000 | 3 | 60,000 | 300 |
| 5,000 | 4 | 400,000 | 2,000 |
| 20,000 | 5 | 2,000,000 | 10,000 |
The point is not the exact figure. It is that a mid size agent deployment reaches enterprise scale spend quickly, and Voice actions at 30 credits push it higher.
Two multipliers turn a modest pilot into a large invoice. The first is actions per request, which rises as agents chain steps and call each other. The second is Data 360 lookups per action. Model both, because the account team forecast usually assumes one action per request and no chaining.
Build the model on peak volume, not average. Agent traffic is spiky, and a rate that looks affordable at the daily mean can breach the credit block on a busy day. A capped rate protects you on exactly those days.
Salesforce sells Agentforce consumption through Pre purchase, Pay as you go, and Pre Commit, and the right choice depends on how confident you are in your volume forecast. The wrong choice can double your effective unit rate.
Pre purchase means you buy a block of credits up front at a discounted unit rate. It suits teams with a credible forecast and a defined pilot to production path. The risk is buying too many credits that then expire unused.
Pay as you go bills credits as you consume them at the list unit rate. It is the simplest to start and the most expensive to run. Use it only for a short discovery window, never as the steady state for production agents.
Pre Commit trades a larger multi year volume commitment for the lowest unit rate. It fits a mature program with predictable, growing agent traffic. The danger is committing to volume you cannot consume, so pair it with rollover and a ramp schedule.
The three buying models at a glance
| Model | Unit rate | Best for | Main risk |
|---|---|---|---|
| Pre purchase | Discounted | Defined pilot to production | Credits expire unused |
| Pay as you go | List, highest | Short discovery only | Runaway steady state cost |
| Pre Commit | Lowest | Mature, growing program | Overcommitting volume |
The pattern in our engagement data is clear. Teams that started on Pay as you go for discovery, then moved to Pre Commit with rollover once volume was proven, paid the lowest blended rate across the term.
The strongest levers all sit before the build, while you still hold leverage. Once agents are in production, the meter has the upper hand.
Negotiate a capped rate per action and per credit, and lock it for the term. A locked rate is the single move that protected buyers most in our engagement data.
Require the production runtime rate, the MCP tool call unit, and the completion pass through to be written as named units with rates. A blank line is a future invoice.
Because the model drives token cost, keep the right to select and switch models. Negotiate credit rollover so unused credits do not expire and force overbuying.
Headless 360 wins where automation and scale matter, and the browser UI still wins where human judgment and low volume make consumption the more expensive path.
High volume, repeatable work with clear rules is where agents pay off. If a task runs thousands of times a day and the logic is stable, headless automation beats human clicks on cost and speed.
Low volume, high judgment work is cheaper on seats. If a task runs a few times a day and needs human context, a licensed seat is far cheaper than metered agent actions.
There is a crossover point where the two paths cost the same. Below it, seats win. Above it, agents win. Finding that point for each workload is the analysis that should drive the design, and it is the analysis the account team rarely runs for you.
Salesforce Headless 360 is an agent first way to consume the Salesforce platform through APIs, MCP tools, and agents rather than the browser user interface. Salesforce announced it on 15 April 2026 at TrailblazerDX. It exposes Data 360, Customer 360, Agentforce, and Slack to external agents and code.
No. As of July 2026 Salesforce has not published a standalone production price list for Headless 360. The billing runs through existing meters, mainly Agentforce Flex Credits and per user add ons, so the cost is real even though a headline price is not posted.
An Agentforce action costs 20 Flex Credits and a Voice action costs 30 credits. Flex Credits are sold at roughly 500 dollars per 100,000 credits, which puts a standard action near 10 cents and a Voice action near 15 cents before any discount.
The undisclosed line items are the production runtime rate for headless agent traffic, the MCP tool call unit, the completion or token pass through, and the rate lock term. Pin each of these as a named unit with a capped rate in the order form before you build.
Flex Credits meter Agentforce actions and Voice today. Headless API calls, MCP tool calls, and model completions can draw on credits or on a separate consumption meter depending on the contract, so a buyer must confirm which meter each traffic type hits.
Salesforce offers three buying models for Agentforce consumption: Pre purchase, Pay as you go, and Pre Commit. Pre purchase and Pre Commit trade a volume commitment for a lower unit rate. Pay as you go is the most expensive per unit and the riskiest at scale.
Yes. The Developer Edition includes a free allowance of 110 requests and 1.5 million tokens per month through 31 May 2026. It is enough to prototype agent flows and MCP tool calls, but it does not model production cost at enterprise volume.
Model the agent traffic in credits first, then negotiate a capped unit rate and a rate lock before you build. The most common mistake we see is teams building on headless agents under an included framing, then discovering the consumption meter at the first true up.
The line items to pin in the order form, the Flex Credit burn model, and the buyer side levers before you go agent first. Independent. Buyer side. Built for procurement leaders.
Join procurement and IT asset leaders who get our benchmarks on Salesforce, SAP, Oracle, Microsoft, and Workday AI pricing. No vendor spin.
