FlowForge · BA/PO Pipeline Proposal

Closing the upstream gap

FlowForge accelerates engineering with 15 agents across 5 steps. The BA phase has 2 agents — both reviewing the spec after you write it. The 6–10 hours before the spec exists — discovery, research, ideation, decomposition, refinement — have no tooling at all.

2
agents on .require today
15
agents on dev pipeline
27
total BA-side agents proposed

The problem

The asymmetry today

Steps 02–06 are built for developers — 15 agents, automated, structured. Step 01 has 2 agents — both post-spec reviewers. The 6–10 hours before the spec gets written have nothing.

2
agents on .require — both post-capture
BA / PO
  • Problem framing none
  • Existing landscape research none
  • Solution option exploration none
  • Spec writing (guided prompts, human-driven) human
  • Gap Reviewer — Explore subagent reviews spec post-capture 1 agent
  • QA Author — generates QA skeleton after approval 1 agent
  • Epic decomposition none
  • Refinement facilitation none
15
agents across 4 steps
Dev
  • .plan — 6 parallel reviewers + Devil's Advocate 6 agents
  • .analyze — 3 parallel gap finders 3 agents
  • .build — TDD pipeline (Builder + Test Validator) 2 agents
  • .verify — 4 compliance checks + Devil's Advocate 4 agents
  • .fix — circuit-breaker repair (max 3 rounds) built-in
  • Express/Fast — full Jira→PR auto-gen auto
  • Traceability chain spec → QA end-to-end

Time audit

Where the time actually goes

A standard feature before it reaches .plan. The spec-capture step has 2 post-capture agents; everything before that has none. The dev pipeline was already at its floor.

Problem framingstakeholder alignment
no tool support
2–4h
Landscape researchJira, Confluence, codebase
no tool support
1–3h
Spec + AC writing/flowforge.require (guided prompts)
human-driven + 2 post-capture agents
2–4h
Spec gap review + QA draftGap Reviewer + QA Author agents
automated
~0.1h
Refinement prepagenda, options, open Qs
no tool support
1–2h
Refinement session+ spec edits
no tool support
1–3h
.plan → .build → .verifyFlowForge dev pipeline
fully automated
0.5–1h
No tool support
Partial support
Fully automated by FlowForge

The proposal

The full pipeline — proposed

Six new commands + one enhanced command extend the BA pipeline upstream of spec writing. Today's 2 agents review the spec after you write it — the proposal adds 25 more agents for everything before and around that moment.

BA / PO zone — upstream
01
/ff.discover
new
Agent-driven interview frames the business problem before any spec writing starts. Forces articulation of success metrics, user impact, and regulatory constraints.
Problem FramerJourney MapperValue ValidatorRisk Spotter
BA / PO0-discovery.md4 agents
02
/ff.research
new
Scans Confluence, Jira, and the codebase — surfaces related epics, prior decisions, established patterns. The spec is grounded in what the system already does.
Confluence ScannerJira LinkerPattern Scout
BA / PO0-research.md3 agents
03
/ff.ideate
new
Generates 3–4 structured solution options — fast-to-ship vs. clean architecture vs. lowest risk. A Recommender agent proposes a preferred option. BA picks one; that anchors the spec.
Option Generator ×3RecommenderLane Estimator
BA / PO0-options.md5 agents
04
/ff.require
enhanced
Spec capture (guided prompts, unchanged) + an expanded review panel. Today's Gap Reviewer + QA Author stay; 6 new specialist agents join them. Catches ambiguity, missing edge cases, GDPR implications before the dev pipeline sees it.
Gap Reviewer ✓ existingQA Author ✓ existingCompleteness AuditorEdge Case HunterAmbiguity DetectorGDPR ProbeStakeholder ImpactScope Guard
BA / PO1-spec.md8 agents (2 existing + 6 new)
05
/ff.split
new
Decomposes epics into independently deliverable stories, identifies the minimum viable slice, marks hard dependencies, suggests lane per story.
Story SlicerDependency MapperMVP Identifier
BA / PO0-split-map.md3 agents
06
/ff.refine
new
Pre-session: structured agenda, open Qs, lane signal, cross-team deps. Post-session: paste in decisions; agent updates spec and Jira automatically.
Agenda BuilderDecision TranscriberJira Sync
BA / POmeeting prep3 agents
07
/ff.prd
new
Assembles discovery, options, spec, lane into a stakeholder-ready PRD — publishes to Confluence automatically. Zero-cost byproduct of the flow.
Document AssemblerConfluence Publisher
BA / POConfluence page2 agents
⇣
Handoff to dev pipeline — only after spec has passed 6-agent BA review. Lane suggestion pre-calculated from ideate + split output.
Dev zone — existing pipeline, unchanged
08
/ff.lane
existing
Auto-detect or assign process lane. Now informed by ideate + split output — lane arrives pre-calculated.
Dev
09
/ff.plan
existing
Technical plan + 6-agent review. Higher-quality spec = fewer plan iterations.
Dev6 agents
10
/ff.analyze
existing
Gap analysis. CRITICAL gaps that previously surfaced here are now caught upstream by the BA review panel.
Dev3 agents
11–13
/ff.build · verify · fix
existing
TDD code generation, spec compliance verification, auto-fix. Unchanged — but operating on a verified, complete spec.
Dev6 agents

Estimated impact

What changes

Grounded in the current team size — 19 FlowForge users, 55 active tickets — observed cycle time, and where time is spent today.

BA / PO time per feature
−45%
Hours saved upstream
Research agent replaces 1–3h of manual Jira/Confluence archaeology. Expanded review panel and ideation reduce rework in refinement. The 2 existing agents already save ~15min on spec quality — the proposal targets the other 6–9h. 3–7h net saved per standard feature.
Spec bounce rate at .analyze
−50%
Fewer critical gaps hitting dev pipeline
The existing Gap Reviewer already catches some spec issues. The 6-agent BA review panel (Completeness, Edge Case, Ambiguity, GDPR, Stakeholder, Scope) catches the rest — before the dev has invested time in .plan. Revised down from an earlier −65% estimate that didn't credit the existing agent.
Refinement meeting duration
−35%
From exploration to decision
/ff.ideate generates 3–4 structured options before the meeting. The session becomes a decision, not an exploration. 90+ min → ~60 min. Revised from −40% — ideation only eliminates option-generation time, not all discussion overhead.
FlowForge adoption reach
2–3×
More people who can drive it
FlowForge is currently a developer tool with a BA entry point that relies on the 2 post-capture agents. With a full upstream pipeline (discover → research → ideate → require → split → refine → prd), BAs and POs initiate and run the entire upstream flow independently. Estimate unchanged — structural shift, not incremental.
Total cycle time (problem → PR)
−33%
Compresses the dominant term
Current: 7–13h BA upstream + 0.5–1h FlowForge dev pipeline. The 2 existing agents save ~15min on spec quality. The proposed pipeline targets the remaining 6–9h. Revised down from −40% to account for partial existing coverage and adoption friction in early phases.
Spec quality floor
↑↑
Higher baseline for .plan input
Today: Gap Reviewer catches spec holes; QA Author drafts test skeleton. Proposed: every spec has also been through discovery, research grounding, and 6-agent review. The .plan reviewers already apply this pattern to technical plans — the proposal replicates it upstream at the requirements layer.

Cycle time

Before vs after

Standard feature, business problem to merged PR. Each segment is a phase. Hover for details.

Today 4–6 days total
~7–14h
2–4h
1–3h
2–4h
✓
2–4h
rework
With BA Pipeline 2.5–4 days total
~4–8h
—
—
1–2h
—
—
No tooling — manual work
Partial / rework
Agent-assisted BA work
FlowForge dev pipeline (automated)
−33%
Revised estimate: 4–6 days → 2.5–4 days total cycle time from business problem to merged PR. The 2 existing agents save ~15min; the proposed pipeline targets the remaining 6–9h of manual upstream work. Revised down from −40% to account for partial existing coverage and expected adoption friction in early phases (discover, ideate). Express and Fast lanes still benefit most: the Jira→PR automation bottleneck is spec quality, not pipeline speed.

Why it compounds

Each phase determines the floor for everything below

The weak link is at the top. The current pipeline is only as good as the spec that enters it.

The quality cascade
BA
Discovery output → spec quality
A spec written without problem framing produces ACs that are technically complete but solve the wrong thing. No agent downstream catches this — only production or sprint review does.
BA
Spec quality → plan iterations
Every ambiguity surfaces as a question during .plan. The 6 plan-review agents flag it, work bounces back to the BA. One cycle = half a dev day wasted before any code is written.
Dev
Plan quality → build correctness
Vague plan → vague tasks → implementation that passes .verify technically correct but doesn't meet business intent. The failure that's hardest to catch.
Dev
The expanded BA panel is a replication of a pattern already in .require
The existing Gap Reviewer already proves that a single spec-review agent catches gaps human review misses. Adding 6 specialist reviewers to that one is the same scale-up .plan already uses. It's not an experiment — it's the same bet, already validated one step upstream.
Where this won't land as predicted
Discovery resistance
/ff.discover depends on BAs trusting an agent-driven interview. First runs will need heavy human correction. Adoption lags the more mechanical commands.
Refinement facilitation
Post-session spec sync only works if the meeting produces structured decisions. Many refinements end with "we'll discuss further" — the tool can't force resolution.
Org dependency
Cycle-time impact depends on BAs adopting FlowForge as their primary workflow. If BAs use it as a final formatting step, the upstream gains don't materialize.
Adoption curve
Commands that ship first (research, require enhancement, prd) validate the model before exploratory ones (discover, ideate) need to be trusted.

Build sequence

Not all 7 commands at once

Start where friction is highest and trust is easiest to build. Each phase validates the model for the next.

Phase 1 — Quick wins
Mechanical · high-trust
/flowforge.research
/flowforge.require (+ BA panel)
/flowforge.prd
Research replaces hours of manual lookup with zero risk. BA review panel adds 6 agents to an existing command. PRD is a zero-cost Confluence byproduct BAs love immediately.
Phase 2 — Refinement layer
Higher leverage · needs trust
/flowforge.split
/flowforge.refine
/flowforge.ideate
Epic decomposition is already done manually — tool just structures it. Refinement meeting prep is easy to validate. Ideation is most useful when framing is already known.
Phase 3 — Full upstream
Transformative · BA-led
/flowforge.discover
Full BA-owned flow
At this point BAs run the entire upstream pipeline independently, handing a verified, research-grounded, agent-reviewed spec to the dev pipeline. FlowForge becomes a full product development tool.