The Agent Development Lifecycle

The complete system for building, running, and improving your AI agents: From writing your first prompt to creating a team of agents, the Agent Development Lifecycle (ADLC) is how enterprises go from AI potential to enterprise performance.

Build any way you want. Deploy on any surface. Trust every outcome.

A designed image of Agentforce Builder with the natural language box at the top and templates below. This image was pulled from the Builder PDP and shows the natural language capabilities.
Phase 1

Design and Build

The foundation of any successful agent deployment is the design and build of the agents. Agentforce Builder is an intuitive workspace that makes AI agents more powerful, predictable, and intentional. Natural-language creation, guided agentic assistance, and Agent Script’s structured logic all work together so you can quickly define the right behavior, troubleshoot issues, and deploy production-grade agents. This gives you all the tools you need so you can scale to more complex, high-value use cases. You can even build agents in your IDE of Choice with the Agentforce Development skill.

A designed image showing batch testing within Agentforce Studio
Phase 2

Test and Evaluate

The Test & Evaluate stage is critical to agent development, and Agentforce Studio gives you everything you need to validate agent behavior before they ever reach the customer. From realistic multi-turn conversation simulation to flexible, custom scoring evals, you can automate testing, simulate conversations at scale, and evaluate probabilistic agent behavior, all without leaving your existing pipeline. You can also batch test your agents with the Agentforce Testing Skill.

The result? Faster iteration, fewer surprises in production, and agents you can trust to behave the right way.

An agent with arrows going to 4 channels showing that the agents you create can be deployed to any of the following channels:  Web Phone Mobile What’s App logo (to represent “other”)
Phase 3

Deploy and Experiment

Putting agents where your customers are is easier than ever before. Agentforce Voice now lets you deliver fast, natural conversations across telephony, mobile, and web channels. Also, you can run controlled experiments on agent configurations in production, comparing instructions, topics, and actions with flexible traffic splits and built-in metrics, so every iteration is driven by real data. This lets you put agents right where your customers are and removes the guesswork.

A designed version of Agent Analytics, pulled from the Observability PDP, that shows performance overview metrics in boxes
Phase 4

Observe and Monitor

The journey doesn’t end with deployment. Agentforce Observability surfaces high-level analytics, engagement trends, conversation-level session insights, and near real-time health monitoring to give you full visibility into how your agents are performing. Additionally, the Session Trace Data Model makes observability fully programmatic, giving teams a complete agent trace that integrates directly into your existing monitoring and DevOps tooling. This creates a continuous feedback loop where every agent interaction becomes a signal for improvement, so you can catch issues fast, optimize confidently, and build lasting trust in your agents at scale.

A designed version of the Salesforce web agent with arrows showing it flow to a help and an event agent. Sub-flows within each include search technical documents, answer support questions, provide session details, and build my agenda. This shows how you can monitor your agents in one place.
Phase 5

Control and Orchestrate

Manage every agent across your organization — regardless of vendor, model, or platform — all in one place. Agent Fabric 2.0 brings deterministic orchestration via Agent Script and expanded multi-cloud agent discovery across Amazon Bedrock, Google Vertex, Copilot Studio, and more. Additionally, a centralized LLM governance and an MCP Bridge makes existing APIs instantly agent-ready without any code changes. This creates a governed, scalable AI landscape where you can eliminate agent sprawl, enforce consistent policies, and confidently orchestrate complex multi-agent workflows across your entire enterprise.

A designed version of Agent Analytics, pulled from the Observability PDP, that shows performance overview metrics in boxes
Headless 360

Build Any Way You Want. Deploy Anywhere.

Salesforce Headless 360 opens every layer of the Salesforce platform — data, business logic, orchestration, and engagement — to any agent, model, or tool, with no need for a browser. The API itself is the UI.

For ADLC practitioners, this means:

  • Build from any IDE, terminal, or coding agent, no Salesforce login needed
  • Invoke Flows, Apex, Data Cloud, and agent logic directly via MCP tools
  • Deploy agents that run headlessly as background workers, collaborating with humans in Slack, email, voice, or any channel
  • Integrate with Claude, Cursor, Databricks, and third-party AI stacks via open standards