Multi-Agentic Pipeline

The multi-agent pipeline processes your business context through a sequential-parallel orchestration of specialized AI agents, producing use cases and (optionally) six scored readiness reports. You choose the route before it starts: Yes - full AI Readiness (~12 min) for use cases grounded in your answers plus 6 reports, or No - use cases only (~8 min) for direct use case generation. Both routes end on the Jamming Board.

The Outcome

You can read the Connection Log during a run and tell whether it is progressing normally, stalled, or finished.

Before You Start

This page is reference for a run that is already underway. Start the run first from Getting Started - nothing here is an action you take.

You do not need to watch the pipeline. It continues server-side if you close the tab, and the Jamming Board is where the output lands either way. Read this when a run looks slow or ends in an error and you need to know which phase it stopped in.

Start Here

Find the newest line in the Connection Log and read the agent prefix in square brackets. That prefix tells you which phase you are in, and the elapsed counter beside it tells you whether it is still moving.

Pipeline Architecture

The pipeline follows this high-level flow:

SYSTEM (init)
  → AGENT: RESEARCH (strategy + initial research, ~2 min)
  → AGENT: RESEARCH (parallel subtopic research, ~2 min)
  → AGENT: WRITER (parallel report generation, ~2.5 min)
  → AGENT: DUAL_PUBLISHER (publishing, ~30s)
  → STEP: REVIEW → Use Case Generator (streaming, ~1 min)

Total wall-clock time: approximately 8 minutes (No path) or 12 minutes (Yes path).

Agent Registry

The pipeline employs 5 distinct agent types. Each has a specific role, operates within defined phases, and produces measurable output.

Execution Phases

Phase 1: Initialization

The system establishes connectivity and plans the research strategy.

[SYSTEM]          → Connecting to use case tool...
[AGENT: RESEARCH] → Planning the research strategy and subtasks...
[SYSTEM]          → Establishing secure connection...

Phase 2: Initial Research

AGENT: RESEARCH begins broad research based on the submitted business context. The agent crawls the company website, analyzes publicly available information, and gathers market intelligence. The Connection Log shows elapsed time counters (10s to 120s) as research progresses.

[AGENT: RESEARCH] → Conducting initial research... (10s elapsed)
[STATUS]          → Research in progress...
[AGENT: RESEARCH] → Conducting initial research... (120s elapsed)

Phase 3: Subtopic Research

The research agent decomposes the broad analysis into specific subtopics and investigates each in parallel.

Typical subtopics for a cloud services company:

• Company's Market Positioning in the cloud/AI ecosystem
• Priority AI Automation Use Cases with quantified ROI potential
• Platform Enhancement Opportunities with competitive differentiation
• Implementation Framework and Guidelines for enterprise adoption

Phase 4: Report Writing

AGENT: WRITER generates both reports from the research data in parallel.

AI Discovery Report

• Structured template with decision matrices, architecture diagrams, scoring tables, and roadmap phases

Deep Research Report

• Narrative document with sections, citations, and data tables covering competitive positioning, industry opportunities, platform strategy, and strategic roadmap

[AGENT: WRITER] → Generating AI Discovery Report... (12s)
[AGENT: WRITER] → Writing decision matrices and scoring tables... (48s)
[AGENT: WRITER] → Composing Deep Research Report... (60s)
[AGENT: WRITER] → Writing research narrative and analysis... (80s)
[AGENT: WRITER] → Finalizing report sections... (96s)

AGENT: DUAL_PUBLISHER then publishes both and confirms with exact character counts:

[AGENT: DUAL_PUBLISHER] → Publishing AI Discovery Report...
[AGENT: DUAL_PUBLISHER] → Publishing Deep Research Report...
[AGENT: DUAL_PUBLISHER] → Reports published (standard: 55,867 chars, structured: 24,087 chars)
[STATUS]                → Multi-agent research completed - preparing for use case generation
[STEP: REVIEW]          → Initiating use case generation from research data...

Phase 5: Use Case Generation

The system streams use cases in real time from the completed research.

• A counter badge increments live: "3 use cases generated" → "6 generated" → "12 generated"
• Each use case is streamed as it is created

[STATUS] → Generating use cases... (3 generated)
[STATUS] → Generating use cases... (6 generated)
[STATUS] → Generating use cases... (9 generated)
[STATUS] → Generating use cases... (12 generated)
[STATUS] → Use case generation complete

Typical output: 10-12 use cases. Each includes:

Results

When generation completes, the pipeline transitions to a success state:

Use cases appear on the Jamming Board ready for review, rating, and collaboration.

Connection Log

Every pipeline event is streamed to the Connection Log in real time. Each entry follows:

[AGENT_TYPE: SUBTYPE] → Message... (elapsed time)

Color coding:

Error Handling & Recovery

If the pipeline encounters an error:

• The Connection Log displays the error message
• The timer pauses
• The Cancel button remains available to abort
• If running in the background, the project is saved with partial results where possible

To retry, return to the AI Assessment intake and start a new session.

WebSocket Resilience

The pipeline is resilient to connection drops:

If your browser tab closes or the connection drops:

1. The backend continues running uninterrupted
2. On reconnect, the frontend polls for completion then re-attaches
3. The tail of the Connection Log is replayed so you see recent output
4. If the run already completed, results load directly from the backend

Continue or start fresh - When you return to an intake with a prior run in progress, you can continue from where it left off or start a completely fresh run.

What Good Looks Like

• The Connection Log keeps producing new lines, and the elapsed counter beside the current phase keeps climbing
• The phases appear in order: RESEARCH, then WRITER, then DUAL_PUBLISHER, then STEP: REVIEW
• DUAL_PUBLISHER confirms both reports published with character counts
• The use case counter streams up to roughly 10-12
• The run ends on "Project Saved" with a project ID in the form company-slug-numeric-id
• Total elapsed time lands near 8 minutes on the No route or 12 minutes on the Yes route

Where This Leads

• Jamming Board - where the generated use cases land for review and rating
• Artifacts - where the reports the WRITER and DUAL_PUBLISHER produced are stored and versioned
• Getting Started - go back to the intake if a run errored and you need to start a new session