Agentforce 360 for practitioners: what's real, what's GA, and what to adopt first

On October 14, 2025, Salesforce unveiled Agentforce 360 at Dreamforce: a new name for the whole stack, a new builder, a new scripting language, native voice agents, and a reimagined Slack. The press release declared it available globally that day — “with continued innovation arriving in pilot and beta over the coming months”. Read that sentence twice. The first half is marketing. The second half is the part you needed to plan around.

Nine months later, the fog has mostly lifted, and the spread of outcomes is instructive. Some announcement-day components matured into GA products that are now the mandatory default — the new Agentforce Builder and Agent Script among them. One flagship, Agentforce Voice, really was GA within a week. And one keynote feature, Slack’s Channel Expert, was discontinued entirely on May 4, 2026, roughly seven months after its debut. GA to dead inside a single release cycle, from the same keynote. That range is exactly why practitioners need status checks, not announcement recaps.

So this is the decode: what Agentforce 360 actually renamed, what each concrete component is, where it stands as of July 2026 against release notes and official docs, and an adopt-now / pilot / wait call for each — ending with the sequence we’d run in a typical service org.

What Agentforce 360 actually renamed — and what it didn’t

Strip the keynote away and Agentforce 360 is an umbrella brand over four things: the Agentforce 360 Platform (the agent runtime, builder, and reasoning engine), Data 360 (Data Cloud, renamed), the Customer 360 apps (Sales, Service, Marketing, Commerce and friends, progressively re-badged with an Agentforce prefix), and Slack as the employee-facing surface. The launch announcement also leaned hard on customer numbers — some 12,000 Agentforce customers, with Reddit credited with deflecting 46% of support cases and cutting average resolution from 8.9 minutes to 1.4. Treat those as vendor-reported launch metrics; the useful signal is that real deployments existed at scale, not the specific percentages.

The renaming deserves one honest paragraph. Data 360 is, by Salesforce Ben’s count, the sixth name for the product previously known as Data Cloud, Genie, and Salesforce CDP before that. Nothing about your existing Data Cloud implementation changed on rename day. Your data streams, identity resolution, and credit consumption carried on as before. If a stakeholder asks whether you need to “migrate to Data 360”, the answer is no — you already run it.

What was genuinely new sat underneath the brand: a rebuilt development experience, a deterministic scripting layer, voice as a first-class channel, and broader model choice — the announcement confirmed Atlas can reason over Anthropic and Google Gemini models alongside OpenAI. If Agentforce itself is still a fuzzy concept in your org, start with our plain-English explainer on what Agentforce is and come back; the rest of this post assumes the basics.

Agentforce Builder and Agent Script: the substance behind the keynote

The most consequential thing announced under the 360 banner wasn’t a brand — it was the rebuilt Agentforce Builder and its scripting language. And it’s also the clearest case study in reading availability labels carefully, because at launch Salesforce’s own developer blog listed the new Agentforce Builder as a pilot, with Agent Script scheduled to enter public beta in November 2025. The open beta of the builder reached everyone in February 2026. GA arrived with the Spring–Summer ‘26 cycle: the release notes now carry a notice titled “The New Agentforce Builder Is Now Generally Available”, and Salesforce’s Summer ‘26 developer guide confirms both Builder and Agent Script as GA. Announcement to GA: roughly seven months.

Now it’s the default, whether you like it or not. Per the same Summer ‘26 guide, from the week of July 13, 2026 the New Agent button disappears from the legacy builder in Setup — new agents can only be created in the new Builder inside Agentforce Studio, and existing agents get a one-click upgrade path. If you’re reading this in the week it publishes, that’s twelve days away.

The builder itself collapses drafting, testing, and deployment into one workspace with three synchronised views — a doc-like editor, a low-code canvas, and a pro-code script view — plus simulation with step-by-step reasoning traces. The script view is where Agent Script lives: a declarative language that mixes prompt text the LLM sees with deterministic logic the runtime executes before the model reasons at all. A minimal example shows the split:

variables:
   is_verified: mutable boolean = False

reasoning:
   instructions: ->
      | Help the customer with their order.
      if @variables.is_verified == False:
         | Verify the customer's identity before discussing any order details.

The if isn’t a suggestion to the model — the runtime evaluates it and assembles the prompt accordingly. That’s the hybrid reasoning idea in one snippet, and it changes what you can guarantee about agent behaviour in regulated workflows. We’ve written a full deep-dive on Agent Script and deterministic agent control, so we won’t repeat it here — but two practitioner notes matter for adoption. First, Salesforce open-sourced the Agent Script toolchain — parser, linter, compiler, and language server — under Apache 2.0, which is an unusually strong commitment signal. Second, the language is still settling cosmetically: the developer guide notes that “topics” were renamed “subagents” in April 2026, and you’ll meet both terms in the docs for a while.

Verdict: adopt now. Not because it’s flawless, but because the legacy builder is being retired underneath you and every new capability lands here first. Migrating a simple agent is quick; migrating a heavily-instructed one is the moment to rewrite your policy rules as deterministic Agent Script rather than porting prompt prose verbatim.

Agentforce Voice: genuinely GA on day one, with telephony strings attached

Voice was the exception to the pilot-and-beta pattern — announced on October 10, 2025 and made widely available on October 21, with Salesforce’s developer blog marking it GA at launch. It gives Agentforce agents a real-time voice channel: low-latency transcription, natural speech synthesis, mid-call actions against CRM data, and warm handoff to a human with the transcript attached.

The strings are infrastructural. Voice rides on Salesforce’s contact-centre telephony layer — Service Cloud Voice lineage, meaning Amazon Connect or partner telephony from the likes of Five9, NiCE, and Vonage. If your telephony already terminates there, adding a voice agent is a configuration project. If it doesn’t, the honest cost of “adopting Agentforce Voice” is a telephony migration with an AI project stapled to it, and you should scope it that way.

Two more caveats from the field. Language coverage launched English-first; the product page tracks current regions and languages, and if you run non-English queues you should verify support there before committing, because coverage has been expanding through 2026 and the docs move faster than blog posts. And pricing is consumption-shaped — Salesforce has described metering by conversation length and tasks completed — so model your call volumes before the pilot, not after. Since launch, Salesforce has kept packaging voice upward into a broader agentic contact centre story, with voice extending into digital channels through the first half of 2026; that extension is newer and we’d let it season.

Verdict: adopt now if you’re on Amazon Connect or partner telephony with English-language queues and a chat agent that already contains well. Pilot otherwise. A voice agent amplifies whatever your text agent already is — including its failure modes, now with a customer listening in real time.

Data 360, Intelligent Context, and Tableau Semantics: the grounding layer

Everything above determines how agents behave. This layer determines what they know, and it’s where GA statuses get genuinely murky — which is itself worth knowing.

Intelligent Context was announced as Data 360’s answer to unstructured grounding: pipelines that extract structure from PDFs, images, and flowcharts so agents can answer from the documents your knowledge actually lives in, configured low-code rather than via a data engineering project. It’s now a documented Data 360 feature, and third-party release trackers record it going GA during the Spring ‘26 cycle — but unlike Builder, Salesforce never gave it a loud, dated GA moment we can point to, and individual pipeline types have moved through beta at different speeds. We’d call the status “GA-with-asterisks”: before you commit a use case to it, check the current release notes for the specific extraction type you need. That ambiguity is a finding, not a footnote.

Tableau Semantics is the other half of grounding — a semantic layer inside Data 360 where metrics, dimensions, and relationships become first-class Salesforce metadata, so an agent asked about “churn” uses the same definition your dashboards do. Salesforce lists it as GA for Tableau Next customers on the Tableau+ SKU and for Data 360 customers, and it requires Data 360 — there’s no standalone path. If your org’s metric definitions currently live in tribal knowledge and competing dashboards, this is real work with real payoff, but it’s a data governance project wearing an AI badge.

The forward signal here is programmatic access. In May 2026 Salesforce shipped a Data 360 MCP server in developer preview, exposing roughly 200 Data 360 API operations to any MCP client, and the Summer ‘26 release made hosted MCP servers — SObject, Data 360, and Tableau — generally available alongside custom MCP server support. Grounding is becoming an interface, not just a pipeline.

Verdict: pilot. Pick one document corpus for Intelligent Context and one metric domain for Tableau Semantics. Both reward starting narrow, and both will punish a big-bang rollout if your source data is untidy.

Slack nine months later: Slackbot won, and Channel Expert is already gone

Dreamforce positioned Slack as the “agentic operating system” — the place humans meet agents. Nine months on, the scoreboard is unambiguous, in both directions.

The winner is the reimagined Slackbot: a personal, context-aware AI agent that went GA on January 13, 2026, rolling out through January and February to Business+ and Enterprise+ plans. It drafts in your voice, briefs you from your channels and calendar, and reaches systems through enterprise search. Alongside it, custom Agentforce agents deploy into Slack channels on any Slack subscription, provided you hold an Agentforce licence — so the agent you built for service can also answer your account teams where they actually work.

The casualty is Channel Expert, the always-on per-channel answer agent demoed at Dreamforce. Slack’s own help documentation now labels it legacy: as of May 4, 2026 it can no longer be added to Slack, with Slackbot recommended in its place. Any enablement, rollout comms, or knowledge-source curation a team invested in it now needs redoing against Slackbot. The feature barely outlived its launch-cycle blog posts.

There’s a durable lesson in that pair. Platform-level capabilities — the builder, the scripting language, the personal assistant — get consolidated into. Point features at the edge of an announcement get consolidated away. When you choose what to build training and process around, weight the platform pieces and treat named point features as provisional until they survive a release cycle or two.

Verdict: adopt Slackbot if you’re on an eligible plan — the setup cost is near zero. Deploy your existing Agentforce agents into Slack where a channel maps cleanly to a queue. Don’t build operational process around any Slack point feature announced within the last two quarters.

The adoption scorecard, and a sequencing plan for a service org

Here’s the whole announcement reduced to the table we wish we’d had in October:

Component	At announcement (Oct 2025)	As of July 2026	Our call
Agentforce Builder (new)	Pilot	GA; legacy New Agent button removed from mid-July 2026	Adopt now
Agent Script / hybrid reasoning	Pilot, beta Nov 2025	GA; toolchain open-sourced (Apache 2.0)	Adopt now
Agentforce Voice	GA (Oct 21, 2025)	GA; digital-channel voice still maturing	Adopt if telephony fits; else pilot
Slackbot (reimagined)	Pilot	GA since Jan 13, 2026 (Business+ / Enterprise+)	Adopt if licensed
Intelligent Context	Announced with Data 360	Documented; GA status murky per pipeline type — verify	Pilot
Tableau Semantics	Announced	GA via Tableau+ SKU / Data 360	Pilot
MCP (hosted + custom servers)	Pilot	GA in Summer ‘26	Pilot
Multi-agent orchestration	—	Beta in Summer ‘26	Wait
Channel Expert (Slack)	Announced	Discontinued May 4, 2026	Moot

For a typical service organisation — a contact centre on Service Cloud, a knowledge base of uneven vintage, agents somewhere between demo and production — we’d sequence it like this:

1. Move to the new Builder and rewrite policy in Agent Script first. The mid-July cutover makes this mandatory anyway; do it deliberately rather than under duress, converting identity gates, refund limits, and disclosure rules from prompt prose into deterministic script as you go.
2. Fix grounding second. Stand up one Intelligent Context pipeline against your highest-traffic document set, and define your top service metrics in Tableau Semantics if you’re already a Data 360 shop. Agents fail on retrieval far more often than on reasoning.
3. Add Voice third, on telephony you already have. Only once the text agent’s containment and escalation numbers are boring. Voice is a channel expansion, not a fresh start.
4. Bring agents to Slack last. Slackbot for individuals, your service agent in channels for swarming and internal deflection. It’s high-visibility, low-risk — a good final act, a poor opening one.

Throughout, gate each stage with the observability tooling that went GA back in November 2025 — the sequencing only works if each step earns the next with measured numbers.

Reading platform announcements like it’s your budget on the line

The nine-month view of Agentforce 360 teaches a repeatable skill, and it’s worth naming because Salesforce has already shown you’ll need it again — TDX in April 2026 brought Headless 360 and Agentforce Vibes 2.0, a fresh wave of announcements now starting their own journey through the same pipeline.

The skill is separating the three layers every keynote compresses into one. There’s the brand layer — Agentforce 360, Data 360 — which changes slide decks and renewal quotes but not your architecture. There’s the platform layer — the builder, the language, the reasoning engine, the semantic model — which arrives in pilot, matures over two or three release cycles, and eventually becomes mandatory. And there’s the feature layer — the named point capabilities demoed on stage — some of which become products and some of which, like Channel Expert, quietly stop existing before your rollout deck is finished.

Time your effort accordingly. Ignore the brand layer beyond updating your vocabulary. Track the platform layer against release notes, because that’s where forced migrations and durable skills live. And let the feature layer prove itself for a couple of cycles before you build process around it. Do that, and announcement season stops being noise and becomes what it should be: an eighteen-month preview of your own backlog, with the GA dates left as an exercise for the practitioner.

Understanding the basics

What is Agentforce 360?

Agentforce 360 is the umbrella brand Salesforce introduced at Dreamforce in October 2025 for its agentic AI stack: the Agentforce 360 Platform (agent builder, Agent Script, and the Atlas reasoning engine), Data 360 (the renamed Data Cloud), the Customer 360 application suite, and Slack. It’s a rebrand plus genuinely new development tooling — existing Agentforce and Data Cloud implementations carried forward unchanged.

Is Agentforce 360 generally available?

The umbrella platform has been GA since October 2025, but components matured at different speeds. Agentforce Voice was GA at launch; the new Agentforce Builder and Agent Script reached GA in the first half of 2026; multi-agent orchestration remains in beta as of Summer ‘26; and Slack’s Channel Expert was discontinued in May 2026. Always verify the specific component against current release notes.

What should a service org adopt first from Agentforce 360?

Start with the new Agentforce Builder and Agent Script, since legacy agent creation is being retired from mid-July 2026 and deterministic scripting is where reliability gains live. Then invest in grounding through Data 360, add Agentforce Voice on existing telephony once text containment is proven, and bring agents into Slack last.

---

Working out which parts of Agentforce 360 belong on your roadmap — and in what order? Talk to us — sequencing this is what we do.