AI Agent Demo Request

Reference these to inform product types, use case mapping, and competitive context when drafting the demo request.

Read EXAMPLES.md for concrete input/output pairs showing correct and incorrect demo request drafts. Study these patterns before researching.

Workflow

Step 0: Inputs and Setup

Determine the company name from $ARGUMENTS or the user's message. Read organizational context files. If a transcript is referenced by name or date, locate it:

!ls -t ~/Library/Mobile\ Documents/iCloud~md~obsidian/Documents/Core\ Vault/Regal/Granola/ 2>/dev/null | head -15 || echo "Obsidian path not found"

If the transcript is pasted directly in chat (not a file reference), analyze it inline during synthesis. Do not spawn a sub-agent for pasted text.

Step 1: Quick Discovery

Run in the main context. No sub-agents. Establish the baseline for targeted dispatch.

1. Salesforce lookup — Run 1-2 SOQL queries via Salesforce-Search:salesforce_query_records:
   • SELECT Id, Name, StageName, Amount, CloseDate, Account.Name, Account.Id, Account.Type, Account.Industry, Account.Website, Owner.Name FROM Opportunity WHERE Account.Name LIKE '%<company>%' ORDER BY CloseDate DESC LIMIT 5
   • SELECT Id, Name, Title, Email FROM Contact WHERE Account.Name LIKE '%<company>%' ORDER BY Name
   • Extract: Opportunity link, Account Industry, Website, contact names, contact emails
2. Obsidian account folder — The obsidian-locator agent (dispatched in Tier 1a) now handles account folder discovery. Do NOT call obsidian_list_files_in_dir in the main context.
3. Granola transcripts — List recent transcripts to identify any mentioning the company.

Capture from Quick Discovery: [COMPANY], [CONTACT_NAMES], [CONTACT_EMAILS], [ACCOUNT_FOLDER_PATH], [SF_OPPORTUNITY_ID], [COMPANY_WEBSITE], [INDUSTRY].

Sub-Agent Output Persistence

Read tools/references/output-persistence.md for the pattern and fallback protocol.

Before dispatching Tier 1 agents, create the output directory:

rm -rf /tmp/ai-demo-request/<company-slug> && mkdir -p /tmp/ai-demo-request/<company-slug>

All tiers write to the same directory. Agent-to-file mapping:

Where <company-slug> is the kebab-cased company name (e.g., oscar-health, camping-world).

Append this block to every agent dispatch prompt, replacing <agent-name> with the agent's subagent_type:

CRITICAL -- Output Persistence: Write your complete, detailed findings to: /tmp/ai-demo-request/<company-slug>/<agent-name>.md Format as markdown with all data, evidence, URLs, and citations. After writing the file, return a 2-3 paragraph summary of key findings only. The file is the primary deliverable. The returned text is a backup summary. If the Write tool fails, return your full findings as text instead.

WRITE EARLY: After your first 2-3 successful research calls, IMMEDIATELY write your current findings to the output file. Do NOT wait until all research is complete. Continue researching and UPDATE the file with additional findings. The initial write is your insurance against output loss. A partial file is infinitely better than no file.

Step 2: Tier 1 Dispatch — Foundation Research

Split into two sub-waves: Tier 1a runs obsidian-locator alongside non-Obsidian agents; Tier 1b dispatches obsidian-analyzer with locator paths.

Tier 1a: Dispatch 3 agents in a SINGLE message

obsidian-locator prompt (max_turns=5):

List all files in Regal/Accounts/[COMPANY]/ and search Regal/Granola/ for transcripts
mentioning [COMPANY]. Return file paths, types (brief, transcript, status, script,
call-flow), and modification dates. If account folder does not exist, report that
and return empty.

CRITICAL -- Output Persistence:
Write your complete, detailed findings to: /tmp/ai-demo-request/<company-slug>/obsidian-locator.md
Format as markdown with all file paths and metadata.
After writing the file, return a 2-3 paragraph summary of key findings only.
The file is the primary deliverable. The returned text is a backup summary.
If the Write tool fails, return your full findings as text instead.

Use the prompt templates in references/research-dispatch.md § Tier 1 for transcript-locator and salesforce-researcher. Populate [COMPANY], [CONTACT_NAMES], and [ACCOUNT_FOLDER_PATH] from Quick Discovery.

Include the Output Persistence block in each agent's dispatch prompt (see Sub-Agent Output Persistence above).

After Tier 1a returns, read obsidian-locator and transcript-locator output. Extract discovered file paths.

Obsidian locator gating:

• If obsidian-locator found no files in the target directories, skip obsidian-analyzer for this account. Note [NO OBSIDIAN DATA: <company>] and proceed to Tier 1 Synthesis.
• If obsidian-locator fails entirely (crash, timeout, or tool_uses: 0), dispatch obsidian-analyzer without path pre-filtering (graceful degradation).

Transcript locator gating: Read transcript-locator output from /tmp/ai-demo-request/<company-slug>/transcript-locator.md. Apply three-state handling:

• Success (locator found transcripts): Dispatch transcript-analyzer in Tier 1b with the discovered transcript paths. Give it max_turns: 8.
• Empty (locator ran but found no transcripts): Skip transcript-analyzer. Note [NO TRANSCRIPT DATA] in synthesis.
• Crash (locator failed, timed out, or tool_uses: 0): Skip transcript-analyzer. Note [TRANSCRIPT LOCATOR FAILED] in synthesis.

Tier 1b: Dispatch obsidian-analyzer + transcript-analyzer (conditional)

Pass the obsidian-locator discovered file paths to obsidian-analyzer's dispatch prompt so it reads only confirmed files instead of searching blind. Pass the transcript-locator discovered paths to transcript-analyzer's dispatch prompt.

Step 3: Tier 1 Synthesis Checkpoint

Wait for all Tier 1 agents to return. Read Tier 1 output files from /tmp/ai-demo-request/<company-slug>/ one at a time, NOT in parallel. A missing file from one agent will cancel all sibling parallel reads (cascade failure). Read them sequentially:

1. obsidian-analyzer.md
2. transcript-locator.md
3. transcript-analyzer.md
4. salesforce-researcher.md

If a file does not exist, note it as "agent produced no output" and continue with the available files. Do NOT let a missing file stop synthesis.

Apply three-tier fallback per tools/references/output-persistence.md.

Synthesize findings into targeting context:

• Primary use case — Identify the single strongest use case across all Tier 1 sources. If multiple use cases are discussed equally, recommend one and note the runner-up.
• Current state — How calls are handled today (manual, IVR, outsourced, etc.)
• Contacts — Confirmed contact names, titles, emails from SF and transcripts
• Existing materials — Scripts, call flows, brand docs found in Obsidian
• Gaps — What is still unknown and needs Tier 2 research

Populate: [PRIMARY_USE_CASE], [CURRENT_STATE_SUMMARY].

Step 4: Tier 2 Dispatch — Contextual Research

Dispatch 2 agents in a SINGLE message. Prompts shaped by Tier 1 findings.

Use the prompt templates in references/research-dispatch.md § Tier 2. Populate [PRIMARY_USE_CASE], [CONTACT_NAMES], [CONTACT_EMAILS], and [CURRENT_STATE_SUMMARY] from Tier 1 synthesis.

Include the Output Persistence block in each Tier 2 agent's dispatch prompt (see Sub-Agent Output Persistence above).

Step 5: Tier 2 Synthesis Checkpoint

Wait for Tier 2 agents to return. Read Tier 2 output files from /tmp/ai-demo-request/<company-slug>/ one at a time, NOT in parallel:

1. slack-researcher.md
2. google-workspace-researcher.md

If a file does not exist, note it as "agent produced no output" and continue. Do NOT let a missing file stop synthesis.

Apply three-tier fallback per tools/references/output-persistence.md.

Update targeting context:

• Confirmed use case — Validated or adjusted based on Slack/email evidence
• Scripts found — Any existing demo scripts, call flows, or agent prompts
• Team context — Builder notes, prior demo attempts, feedback from Slack
• Brand materials — Voice preferences, product naming from email correspondence
• Targeted questions — Specific gaps that external research should address

Populate: [TARGETED_QUESTIONS].

Step 6: Tier 3 Dispatch — External Research

Dispatch 2 agents in a SINGLE message. Targeted by Tier 1+2 findings.

Use the prompt templates in references/research-dispatch.md § Tier 3. Populate [PRIMARY_USE_CASE], [CURRENT_STATE_SUMMARY], [TARGETED_QUESTIONS], [COMPANY_WEBSITE], and [INDUSTRY] from Tier 1+2 synthesis.

Include the Output Persistence block in each Tier 3 agent's dispatch prompt (see Sub-Agent Output Persistence above).

• deep-research-agent -- Conditional. Dispatch when: (a) the prospect is in a complex industry (healthcare, financial services, insurance) where regulatory/operational nuance affects the demo request, or (b) Tier 1-2 findings reveal thin account context (no prior transcripts, no existing account brief). Read tools/references/deep-research-dispatch.md for query construction. Give it max_turns: 8.

Step 6.5: Demoability Filter

Before drafting, classify every capability found across all tiers. This is the critical translation step: converting raw prospect requirements into demo-ready tasks.

For each capability the prospect mentioned (from transcripts, Slack, emails, or call notes), classify it:

Output of this step: A filtered, prioritized list of core tasks ready for Part 1, with reframing notes for any "demoable with reframing" items. Journey-level and production-only items are separated and placed appropriately.

Consult references/demo-scripting.md § Dynamic Variables & Platform Capabilities for the full variable syntax and action inventory.

Step 7: Final Synthesis and Demo Brief

Wait for all Tier 3 agents to return. Read Tier 3 output files from /tmp/ai-demo-request/<company-slug>/ one at a time, NOT in parallel:

1. company-researcher.md
2. web-search-researcher.md
3. deep-research-agent.md (only if deep-research-agent was dispatched in Step 6)

If a file does not exist, note it and continue. Do NOT let a missing file stop synthesis.

Apply three-tier fallback per tools/references/output-persistence.md.

Draft the Demo Brief in two parts using the filtered task list from Step 6.5.

Consult references/form-fields.md for field definitions and best practices. Consult references/demo-scripting.md for demo strategy best practices. Consult references/examples.md for quality benchmarks.

Synthesis quality checks (verify before presenting the draft):

1. Every numbered core task uses only pre-built actions or dynamic variables. No task depends on custom actions or live API calls unless the builder confirmed they will configure them.
2. Any capability that was reframed (CRM lookup to contact property, API call to verbal confirmation) uses the correct {{contact.customProperties.*}} or {{task.*}} syntax in the Script field.
3. Journey-level items (retry cadence, trigger conditions, multi-channel sequences) appear only in the Use Case context sentence, never as numbered core tasks.
4. Production-only items appear only in builder notes with the "Production:" prefix.
5. The Script field references specific dynamic variables by name (e.g., {{contact.customProperties.medication}}), not abstract placeholders like "[patient data]".

Output Format — Part 1: The Form

Present the form draft in this exact format:

## AI Agent Demo Request Draft — [Company Name]

**Brand:** [Company name]

**Customer Industry:** [From SF Account.Industry or infer from company context]

**Company Website Link:** [https://www.example.com — from SF Account.Website or research]

**New or Existing Customer:** [New / Existing]

**Inbound or Outbound:** [Inbound / Outbound]

**Type of Agent:** [Comma-separated from valid options]

**Use Case/Role of Agent:**
[Company] is a [what they do, where]. The AI agent handles [inbound/outbound] [calls] for [purpose/when it activates].

Core tasks:

1. [Demoable capability] — [technical detail: routing examples, API mechanisms, dialogue snippets]
2. [Demoable capability] — [detail]
3. [Demoable capability] — [detail]
...
N. Transfer to human — [specific conditions]

[1-3 lines max: only voice/tone preferences, prior demo references ("similar to hear.com"), or build-critical requirements (voice clone, bilingual). Omit if none.]

**Script or Recordings in Regal:**

Flow:
- [Intro path]
- [Key path 1: brief description]
- [Key path 2: brief description]
- [Transfer/escalation conditions]

Knowledge base: [1-2 parent URLs — builder will explore subpages]

**Measuring Success:** [Primary metric — keep to a few words]

**# Answered Calls per Month:** [Number or blank]

**Requested by Date:** [YYYY-MM-DD — default to 3 business days from today]

**Urgent Justification:** [Only if < 3 business days]

**Salesforce Opportunity Link:** [Full URL from SF research]

Output Format — Part 2: Demo Strategy

Append the Demo Strategy after Part 1. This gives the builder deeper context for configuring the agent.

Reference references/demo-scripting.md for best practices on each section.

Current State

Describe how the company handles this use case today. Source from Tier 1 transcripts, Obsidian notes, and SF data. Include: current process, pain points, call volume, technology in use.

Demo Approach

Recommend the demo call flow structure. Map to the 6-stage framework in demo-scripting.md (Intro, Verification, Core Task, Objection Handling, Escalation, Close). Specify what the builder needs to configure at each stage.

Challenge Scenarios

Select 3-5 challenge scenarios relevant to the use case type from demo-scripting.md. Adapt each scenario to the company's specific context using Tier 1-3 findings. Present as a table with Scenario, Challenge, and Response Approach columns.

Brand and Voice

Synthesize brand voice from all research tiers: website copy style, email tone, transcript dialogue patterns, product naming. Also reference brand-guidelines.md for Regal's own brand standard (confident, enterprise-ready, intentional) to ensure the demo request framing aligns with how Regal presents itself. Provide an example greeting line the builder can use. Reference demo-scripting.md § Brand Alignment.

Supporting Materials Found

List all materials discovered during research that the builder should reference:

• Scripts, call flows, or agent prompts from Obsidian or Drive
• Knowledge base URLs (max 2-3 parent pages, annotated)
• Brand guidelines or voice documentation
• Prior demo references or recordings

Save to Obsidian

After drafting the Demo Brief, save the full output (Part 1 + Part 2) to the account's Obsidian folder:

Path: Regal/Accounts/[COMPANY]/AI-Demo-Request - [YYYY-MM-DD] - [COMPANY].md

Use obsidian_append_content to create the file. If the account folder does not exist, create it by writing to the path (Obsidian MCP will create intermediate directories).

The saved file should contain both Part 1 (the form) and Part 2 (demo strategy) as a single markdown document. Add a YAML frontmatter block:

---