Build Agent with AI
AI funding and usage
AI included uses REDA credits. Bring your own AI uses your organisation’s provider account, billed directly by that provider. The licence selects one funding model for the organisation. In either model, REDA AI governs configured tools and actions in Salesforce. Set up your licence and AI accounts.
In customer-account mode, the selected model determines which Deep Reasoning levels and capabilities are available. Review the AI Account selector’s supported levels and warnings; do not assume every provider supports the same combination. Admin AI tools and document processing use the organisation default connection. Configure the default and test the workflow.
What It Does
The feature is an analyser, not a generator. It works on an existing agent definition (with at least a display name and persona) and proposes targeted edits in five categories:
| Category | What gets recommended |
|---|---|
| Persona | A rewritten persona that’s more specific about scope, tone, target user and what the agent should not do. |
| Token Budget | A suggested budget that fits the agent’s actual prompt and tool footprint — up if the agent is bumping its limits, down if there’s headroom to spare. |
| Tools | Built-in or custom tools that the agent should add or drop based on its purpose. |
| Context Fields | Object × field configurations that would give the agent better record context, framed as additions, removals or label improvements. |
| Suggestion Chips | Quick-action chips for the chat empty state, sized and worded to match the agent’s purpose. |
Suggestions for Persona and Prompt Complexity Budget overwrite their target fields when applied. Suggestions for Custom Prompts, Context Fields and Suggestion Chips create new records linked to the agent — existing records aren’t modified.
How It Differs from Related Features
- The Context Field Optimizer is a focused, per-record optimiser for one Context Field record at a time. Build Agent with AI is broader — it surfaces context-field opportunities across the agent alongside everything else.
- Agent Runtime Preview is read-only and free; it shows what the agent will load. Build Agent with AI is a model-driven analyser that proposes changes; it costs credits.
The three features compose well: preview to see the current state, run Build Agent with AI for a broad pass, then run the Context Field Optimizer per record for finer tuning.
How a Run Works
-
Open the agent
The agent should have at least a Display Name and a Persona. The more specific the persona, the more useful the suggestions.
-
(Optional) Add guidance and a sample file
Enter free-text instructions (up to 15,000 characters) describing what the agent should do — e.g. “reads supplier invoices and updates the matching Purchase Order.” This is the primary directive the suggestions align to. You can also attach a representative sample file (PDF, image, CSV, TXT, or JSON) the agent will process at runtime; the AI reads it to produce concrete extraction and context-field guidance tailored to that document, without copying its contents into the prompt.
-
Click Build with AI
The analyser reads the full agent state — persona, channel, prompts, tools, context fields and suggestion chips — alongside the relevant object schema in your org.
-
Review the proposals
Suggestions are grouped by category. Each item shows the rationale and the exact change that will be applied.
-
Apply, edit or skip
For each suggestion you can apply as-is, edit before applying, or skip. Nothing changes until you save.
-
Save
Applied suggestions are committed to the agent: persona and token budget overwrite, prompts/context-fields/chips are added as new records.
Credit & Time Cost
A run makes two AI calls under a single token budget — the first covers the core analysis (persona, settings, prompt suggestions), the second the context-field and suggestion-chip pass. The credit cost scales with the size of the input (mainly the agent’s current configuration plus the relevant schema) and the size of the output. The platform shows the actual credits consumed after the run.
If a call times out, the platform applies a fixed timeout penalty of 20 credits and returns an error you can retry.
See Credit Usage Optimisation for the wider cost model.
When to Run It
- After creating an agent. The fastest way to turn a name + persona into a working configuration with prompts, context and chips.
- After meaningful changes. A new persona, a new channel, or significantly different tools all change what the agent should know — rerun to catch the gaps.
- After schema changes. New objects or fields in your org create context-field opportunities the analyser can surface.
- When trimming costs. Suggestions for unused tools and oversized token budgets are concrete handles for reducing credit consumption.
What It Does Not Do
- It doesn’t create the agent definition itself — you create the agent first, set a persona, then run the analyser.
- It doesn’t auto-apply. Recommendations are advisory; nothing changes without your save.
- It doesn’t enforce field-level security at design time. Field-level security is enforced at runtime when the agent actually reads context.
- It doesn’t replace your judgement on tone, brand voice, or domain-specific guardrails.
Best Practices
Make the persona specific before running. “Help users with their work” produces generic suggestions. “Triage inbound product feedback and route it to the right product manager” produces sharp ones.
Read the rationale, not just the diff. The rationale often surfaces something you’d want to change about the agent itself, not just the suggestion at hand.
Apply selectively. Accepting every suggestion is usually worse than accepting half of them carefully. The analyser optimises for breadth; you optimise for fit.
Verify in the Runtime Preview. After applying, open Agent Runtime Preview to confirm that the new prompts, tools and chips are loading the way you intended.
Frequently Asked Questions
Yes — it suggests tools to add or drop based on the agent’s purpose, alongside persona, token budget, context fields and suggestion chips.
No. Custom Prompt suggestions are additions — new records linked to the agent. Persona and Prompt Complexity Budget suggestions overwrite their fields when applied, but you can always revert manually.
Yes. Subsequent runs typically produce fewer suggestions as you accumulate accepted improvements. There’s no rate limit beyond your credit budget.
Use Edit to adjust the suggestion before applying, or Skip to drop it entirely. Applied items remain editable on the agent definition afterwards.
Build Agent with AI runs across the whole agent at once and proposes additions to context fields among many other categories. The Context Field Optimizer focuses on a single Context Field record at a time, fine-tuning its four configuration values. Use Build Agent with AI for breadth; use the Optimizer for depth on a specific record.
Typically seconds. Larger agent definitions and orgs with many fields on related objects take a little longer because more schema is sent to the model.
Next Steps
Open the Agent Runtime Preview to verify the post-apply state, or read the Context Field Optimizer for per-record context-field tuning.