Enhance Your Agents in the New Agentforce Builder
More consistent agent behavior on every run
More consistent agent behavior on every run
The new Agentforce Builder in Agentforce Studio is a authoring workspace built around the agent development lifecycle — configuration, execution traceability, context management, and deployment.
At its core is Agent Script, a new scripting language that combines the creative problem-solving of LLMs with deterministic logic that follows your business rules. This means your agents adhere to your business rules every time, not just most of the time.
If you’re building or running Agentforce agents today, the new builder is for you. Starting the week of July 13, 2026, agent creation will move exclusively to the new builder. Agents already built on the legacy builder will continue to work and remain editable, but no new agents can be created there.
It's probably one of the bigger joys to know that what you think is going to happen actually happened. That’s where we are with Agent Script. It really takes away the frustration because you know, without a shadow of a doubt, it’s going to run 100% of the time every time.
Grant RobersonAgentforce Admin, Datasite
Conversations with teams that have already converted consistently highlight three main benefits.
The new builder lets you combine deterministic rules with LLM reasoning through Agent Script. Hybrid reasoning is unique to the new builder. It holds your agent to your business logic exactly as written, even on regulated or high-stakes flows. Agent Script also reduces latency by letting the agent follow structured logic instead of re-evaluating every topic and action at every turn.
After rebuilding its service agent in the new builder, Datasite’scase deflection climbed from the low-60s to 82%.
Agentforce Observability is built in. See a summary of your agent’s reasoning for every message. Dig into granular session traces — including searchable spans, variable changes, and timestamps — to pinpoint root causes and make informed fixes.
Xero rebuilt its Customer Service Agent on Agent Script and is now scoring every live session against brand guidelines using new custom scorers (in beta). The new agent resolves about 60% of queries instantly across customer service, lead outreach, lead nurture, and employee support in Slack.
This will be the exclusive home for all future enhancements. That includes Agentforce Voice, expanded language support (25+ languages beyond English and French), upgraded speech models (Deepgram Nova3 STT, ElevenLabs v3 TTS), per-subagent model selection, and agent-to-agent orchestration.
Built on Agentforce Voice with Agent Script, Zing Health’s agent, Mia, fields 150,000+ member calls a month. It resolves intricate benefits and drug coverage questions across 33 plans — delivering instant, bilingual (EN/ES) support with intelligent routing and seamless handoff to human agents.
After pivoting off a legacy AI tool, Workday rebuilt on Agentforce with a two-org architecture and Agent Script. It set a 10% case deflection target for the year and hit 21% by end of Q1 — more than doubling it — with faster iteration, more deterministic control, and measurable results in weeks.
| Capability | New Agentforce Builder with Agent Script | Legacy Builder |
|---|---|---|
| Reliability and reasoning | Hybrid reasoning: Combines deterministic rules + LLM reasoning. Reduced latency. | Reasoning with natural-language prompts only. Limited guardrails, higher latency. |
| Testing and debugging | Per-message reasoning summary. Granular session traces with searchable spans, variable changes, timestamps, and text + code for root-cause analysis. | Conversation previews with limited diagnostics on the plan canvas. No granular span-level detail. |
| How you build | Build your way: natural language in Canvas view, write directly in Script view, or use VS Code, CLI, or AI coding tools like Agentforce Vibes, Claude Code, and Cursor. Chat with Agentforce for guidance at any experience level. | UI experience only. No built-in AI assistance. |
| Voice and new channels | Agentforce Voice and all future voice enhancements ship here, including expanded language support. | Existing voice agents keep working; no new voice features. New voice agent creation stops the week of July 13, 2026. |
| Observability | New observability features land here first, including custom scorers (beta). | Supports Agentforce Observability today; may not support new features. |
| Workspace | Embedded in Agentforce Studio: build, test, and monitor in one place. Builders don’t need admin-level permissions. | Managed in Salesforce Setup across many tabs. Requires admin-level permissions. |
| Model selection per subagent | Choose the LLM at the subagent level to fine-tune which model handles which task. | No per-subagent model selection. |
| Agent-to-agent orchestration | Multi-agent orchestration with Agent Script: a single "Superagent" front door coordinates expert subagents behind the scenes, preserving shared context. | No agent-to-agent orchestration. |
| Future enhancements | All future Agentforce innovation. | No new features or enhancements planned. |
| Where you are today | Start here |
|---|---|
| Haven’t built an agent yet | Start in the new builder. New agent creation moved to the new builder only the week of July 13, 2026. |
| Started in the legacy builder, not yet in production | This is your lowest-risk window to switch. Move now, before testing cycles and dependencies stack up. |
| Already in production | Plan a staged enhancement project. Pick one agent, set a baseline, and stage the work. Rebuild in the new builder, fix regressions, then harden with Agent Script. Test after each batch. Datasite did exactly this and saw case deflection climb from the low-60s to 82%. |
| Agent is outdated or no longer in use | Skip the rebuild. If you need a replacement, start fresh in the new builder. |
| Just exploring Agentforce | Try the new builder hands on in Agentforce Labs — no Salesforce org required. |
Start with the Agentforce Migration Guide. It covers planning, prerequisites, and readiness checks so you can scope the work before you touch either path below.
From there, choose the path that fits how you like to work, or combine them. For example, you could use the Agentforce conversion path to stand up a baseline quickly, then iterate with development skills. To get the full benefit of Agent Script, review and refine your agent's configuration after the upgrade and before going live.
Once you activate your rebuilt agent in the new builder, your original legacy agent is automatically deactivated. You can leave it as-is, archive it, or delete it.
| Path | Choose this if… |
|---|---|
| Convert with Agentforce (UI-driven) | You want a new version that closely mirrors your original agent, so you can compare, diff, and validate side by side. |
| Salesforce development skills (CLI + AI coding tools) | You want a redesigned agent built around new patterns from the ground up — using AI coding tools or your own development workflow to author and iterate. |
Agentforce Migration Guide: Official help documentation
Upgrade Your Agent to the New Agentforce Builder: Step-by-step instructions for the in-Studio upgrade flow
Agentforce Developer Guide: Common Patterns: See what Agent Script unlocks
Get to Know the New Agentforce Builder (Trailhead): Free, hands-on learning path
Build Enterprise-Ready Agents with the New Agentforce Builder: Architecture and best practices for production-grade agents
Agentforce Labs: Build with Agent Script in a sandbox
Your existing agents built in the legacy builder will continue to run and remain editable — they are not being shut down. However, as of of July 13, 2026, you're no longer be able to create new agents in the legacy builder. All new agent creation moves to the new Agentforce Builder. No new features or enhancements are planned for the legacy builder, so teams that want access to Agent Script, Agentforce Voice, per-subagent model selection, multi-agent orchestration, and other future capabilities will need to rebuild in the new builder.
Agent Script is a new scripting language built into the new Agentforce Builder that combines LLM reasoning with deterministic logic. In the legacy builder, agents interpret natural-language instructions on every turn — which means behavior can vary, especially on regulated or high-stakes flows. Agent Script lets you define exactly which steps your agent must follow and where it can reason freely, so your business rules execute every time. It also reduces latency by letting the agent follow structured logic instead of re-evaluating every topic and action on every turn.
No. When you use the in-Studio upgrade, a new draft version of your agent is created in the new builder — your original legacy agent stays intact and keeps running. You can compare, test, and validate the new version before making it active. Once you activate the rebuilt agent in the new builder, your original legacy agent is automatically deactivated. You can leave it as-is, archive it, or delete it.
The new Agentforce Builder is a workspace built around the full agent development lifecycle — configuration, execution traceability, context management, and deployment. Key differences: (1) Agent Script support for deterministic, reliable agent behavior; (2) built-in Agentforce Observability with per-message reasoning summaries and granular session traces; (3) flexible build options including Canvas view, Script view, VS Code, CLI, and AI coding tools like Claude Code and Cursor; (4) a unified studio so builders don't need admin-level permissions; and (5) all future Agentforce innovation — Agentforce Voice, per-subagent model selection, multi-agent orchestration — ships here exclusively.
There are two paths, and you can combine them. The first is the in-Studio upgrade: in Agentforce Builder, specify the agent and version you want to upgrade. Agentforce automatically creates a draft version in the new builder with your original subagents, actions, system messages, settings, data, and connections — converted to Agent Script. This works well if you want a new version that closely mirrors your original so you can compare, diff, and validate side by side. The second path uses Salesforce development skills — the CLI and AI coding tools like Claude Code or Cursor — to rebuild from the ground up with new Agent Script patterns. Choose this if you want to redesign your agent's architecture rather than replicate it. To get the full benefit of Agent Script, review and refine your agent's configuration after the upgrade and before going live.
It depends on your current situation:
Take a closer look at how agent building works in our library.
Launch Agentforce with speed, confidence, and ROI you can measure.
Tell us about your business needs, and we’ll help you find answers.