A comparison between Alchemist AI Pro™ and managing requirements through traditional means or traditional requirements software solutions.
Alchemist AI Pro™ · Slide 1 of 3
Without vs. With Alchemist AI Pro™
The Manual BA
Whole documents arrive unstructured; objects, processes, and actors hidden inside prose; SMEs, if available, spend cycles re-explaining context.
Alchemy-Assisted BA
Reading code repositories directly, consuming DB definitions, leveraging user guides, SOPs, and other reference material.
Existing-system contracts, interface response times, and upstream/downstream context are captured ad hoc — if at all.
Domain knowledge injected via proprietary ontology; existing systems tagged as system contracts with measurable interface SLAs.
Analysts jump from the ask straight to epics and stories — the 'how' contaminates the 'what,' and user value is diluted.
Conversation locked to high-level features only — the user's WHAT before any HOW (Working Backwards / PR-FAQ discipline).
Unstructured interviews; tacit assumptions slip through; ambiguity is discovered late — typically by developers, mid-sprint.
Socratic dialogue drives use cases out of features; ambiguity, vagueness, and tacit assumptions surfaced before dev hand-off.
Whiteboards photographed and lost; word meanings commingle; capabilities and processes are conflated in the model.
Capability ontology explicitly separated; new actor / object / process declared on demand and tagged in the model.
Free-form acceptance criteria; the eight named requirements defects (ambiguity, vagueness, complexity, etc.) persist into dev.
EARS-formatted requirements eliminate the eight named defects; backlog priority captured at SME signoff for dev-ready turnover.
Alchemist AI Pro™ · Slide 2 of 3
Without vs. With Alchemy Crew
Vibe Coding
Prompts to Claude Code, Kiro, and Codex produce plausible-looking code with no scope baseline; estimates are gut-feel; "done" is undefined. Scope creep is invisible until it ships broken.
Alchemy-Assisted Dev
EARS-validated stories drive sizing; cross-feature dependencies traced before sprint start; every estimate is backed by structured acceptance criteria, not a prompt guess.
Each Claude Code / Kiro / Codex session is siloed - no shared data model, no API contracts, no cross-feature vocabulary. Schemas generated today contradict what was scaffolded last sprint.
Live ontology - actors, objects, processes, rules - declared once and referenced everywhere; cross-feature consistency enforced by Alchemy Crew before code is written.
UI direction is "make it look like X" - no persona grounding, no test-case derivation, no acceptance baseline. MVPs ship passing vibes, not validated user journeys.
Working Backwards locks the WHAT before any HOW; test cases derived from use cases, not invented post-hoc; acceptance criteria precede the first prompt to the AI tool.
Codex / Claude Code scaffold whatever schema the next prompt implies; routes accumulate without data contracts; migrations pile up as the model evolves via more vibe sessions.
Routes and schema derived from validated system contracts with measurable interface SLAs; every table and endpoint traces back to the requirement that demanded it.
Vibe coding collapses domain vocabulary - "user," "customer," "profile" used interchangeably across AI sessions; business rules are buried in generated code, invisible and untestable.
Canonical capability ontology enforces naming discipline; business rules are declared as first-class artifacts before code generation, not reverse-engineered from AI output after the fact.
Stories are ad hoc prompts - "build a login flow" - with no acceptance criteria; Claude Code / Kiro / Codex ship passing tests against undefined expectations; "done" means it compiled once.
EARS-formatted AC generated from Alchemy Crew's spec pipeline; every commit tagged to its spec, test case, and use case - full traceability from requirement to merge.
Alchemist AI Pro™ · Slide 3 of 3
Without vs. With Alchemy Away Team
Ad Hoc Patching
Fixes target the first error in the log; AI tools patch the visible symptom without anomaly classification; regressions compound because origin and impact are never separated.
Guardrail-Driven Repair
The Away Team classifies each anomaly by origin and impact before writing a single line; diagnostics, UX empathy, and architecture are all examined - fixes trace back to the spec that defined correct behavior.
AI debug sessions start cold with no memory of architectural intent; the LLM reasons backward from current behavior - complexity accumulates with every patch, exactly as Lehman's Law predicts.
The Away Team reads Tasks, Specs, and Tests before reading code; full cross-feature context is held; complexity is actively measured and managed - fixes are driven by original intent, not current behavior.
UX is limited to the happy path; HTTP 200 = done; error states, empty states, RBAC boundaries, and exception flows are discovered by users in production, not by the team before deploy.
The Away Team maps effectiveness, efficiency, and satisfaction across the entire journey - sidetracks, edge cases, and exception paths included; ISO 9241-11 usability criteria are exercised, not assumed.
Route and DB fixes are applied in isolation; no signal on downstream impact; latency spikes and error-rate regressions surface in production because no cross-layer signal was monitored before deploy.
The Away Team monitors the four golden signals - latency, traffic, errors, saturation - across all service layers; root cause identified before the patch is written; all contracts validated before merge.
RBAC is patched as if-else blocks; role boundaries are implicit and untested; a single vibe prompt can silently break access for an entire user class with no guardrail to catch it.
The Away Team treats roles as declared first-class actors; every route is mapped to its authorized role set per the Ferraiolo-Kuhn RBAC model; reuse and exception patterns are explicit, tested, and documented before merge.
Developers read the code to understand the bug; the fix is whatever makes the failing test green - no reference to original specification; every patch adds silent undocumented assumptions to the system.
Cleanroom discipline: the Away Team reads the spec before the code; correctness is verified against original intent; a repair that deviates from spec triggers a spec update first, not a code update.
About the Author
Jamie Campbell
Technology Leader · ACC3 International
With over 25 years of enterprise application experience, Jamie Campbell is a digital transformation expert who specializes in driving end-to-end SDLC modernization. He acts as a strategic agent of change, leveraging best-in-class digital process automation and Agile methodologies to deliver major business transformations.