The dates you need first
Winter ’27 is rolling out now, and the calendar has one trap in it.
| Milestone | Date |
|---|---|
| Release notes published | 19 August 2026 |
| Sandbox preview opens | 28 August 2026 |
| Production wave 1 | 4 September 2026 |
| Production wave 2 | 2 October 2026 |
| Production wave 3 | 9 October 2026 |
The trap is the gap between preview and the first production wave: it is one week, not the six weeks many teams assume by default. If your instance is in the first group, your testing runway is dramatically shorter than the planning habit most admins have built up. Check which wave your instance falls into on the Salesforce Trust status page rather than assuming.
Sales Cloud is now Agentforce Sales
If the name change catches you off guard, it has been in motion for a while — Salesforce rebranded Sales Cloud to Agentforce Sales at Dreamforce ’25, and it was formalized in the Spring ’26 release. It is the same product with a new name and a heavier emphasis on agents.
Practically, this matters for documentation, internal training material, and any runbooks that reference the old name. It is a rename, not a migration — nothing breaks.
One genuinely useful change in cadence: Sales features now ship as often as monthly rather than only at release boundaries. Your change management process needs to account for that. A quarterly review of "what's new" is no longer sufficient.
The new Agentforce Builder and Agent Script
This is the most consequential change for teams actively building.
Agentforce Engagement agents can now be created in the new Agentforce Builder inside Agentforce Studio, which is built on Agent Script — a declarative way to express business rules and guardrails. Agents built with Agent Script use the new Builder; legacy agents continue in the older Builder in Setup.
That split is the thing to plan around. You will be operating two builders simultaneously for some period, and that has real consequences:
- Your team needs to know which builder a given agent lives in before touching it
- Documentation and runbooks need to specify the builder, not just the agent
- Migration is a project to scope deliberately, not something to do opportunistically mid-sprint
Our recommendation: build new agents in the new Builder, leave working legacy agents where they are, and schedule migration as its own piece of work once you have real experience with Agent Script.
Observability and Custom Scorers
Agentforce Observability gives you a way to track agent performance and see where agents are failing. Alongside it, Custom Scorers (in beta) let a team define its own evaluation logic for an agent session — not just Salesforce's built-in quality metrics, but a pass/fail question tied to what your specific use case actually needs.
This is the feature we would prioritize above everything else in this release, and it is the one most likely to be ignored.
Built-in quality metrics answer "did the agent respond coherently." Custom Scorers let you answer the question you actually care about — did it qualify this lead against our criteria, did it escalate when our policy requires escalation, did it avoid the specific failure mode that burned us in month two. That is the difference between monitoring and evaluation.
Note it is in beta. Pilot it; do not make it a compliance control yet.
Agentforce Prospecting
Several practical additions:
- Scheduling options for agent runs
- Prioritizing recently engaged contacts
- Researching accounts outside a dedicated prospecting group
- An Agent Activity view
The scheduling and Agent Activity additions are the ones with immediate operational value. Being able to see what the agent has been doing, in one place, closes a visibility gap that has generated a lot of "is this thing even working" conversations.
Agentforce Voice
Sharper transcription and speech that sounds less robotic. Unglamorous, and genuinely important — transcription accuracy is the ceiling on everything a voice agent does downstream. If you evaluated Voice previously and rejected it on quality, it is worth re-testing rather than trusting your prior conclusion.
Einstein Conversation Insights
- Suggested next steps surfaced once meeting transcripts become available
- Post-Meeting Voice Notes capture details from conversations that were never transcribed
The second one quietly closes a real gap. A large share of what actually moves a deal happens in conversations no system ever recorded — hallway calls, quick check-ins, the debrief in the car afterwards.
Service Cloud
Agentforce self-service portals, plus IT Service capabilities including hardware lifecycle management and a CMDB conversational agent. If you run internal IT support on Salesforce, the CMDB agent is worth a close look — configuration lookups are exactly the high-volume, low-judgment queries agents handle well.
A pragmatic upgrade checklist
- Confirm your production wave first. Everything else depends on whether you have one week or five.
- Regression-test existing agents in the preview sandbox. Specifically test escalation paths and guardrails — those are where behavior changes hurt most.
- Do not migrate agents to the new Builder during the upgrade window. Two variables at once makes any problem twice as hard to diagnose.
- Pilot Custom Scorers on one agent. Pick your highest-volume use case and write one honest pass/fail question.
- Re-test Voice if you rejected it before.
- Update runbooks for the Agentforce Sales naming and for the monthly Sales release cadence.
Release windows are where under-governed agent deployments tend to surface their problems. If you want a second set of eyes on your regression plan — or you are looking at this release and realizing you have no evaluation baseline to regress against — that is the kind of engagement Maple Wave AI is built for.