Reverse Engineering
Assisted by AI
Transformation
Salesforce
Removing the illusion of continuous living documentation to build a As-Is foundation based on evidence — deterministic, traceable and validated by humans.
Discovery and Preservation
In ecosystems Salesforce with years of organic evolution, vital rules inhabit exclusively in thousands of lines of Apex, fragmented Flows and in the minds of few experts. The Purpose It's not implement automated ad aeternum documentation.
Freeze As-Is
Execute an intensive phase of discovery to map and understand exactly the system behavior in its current state.
Reveal the Hidden
Extract embedded rules into the code, avoiding loss of critical business knowledge during decommissioning or refactoring.
Base Engineering
Provide reliable inputs for architects and product owners to consciously decide the which should be preserved, redesigned or discontinued.
Discovery guided by Critical Days and Risk (ex. Lead-to-Cash). AI agents follow the execution trail (call stack), mapping all components — quickly isolating dead code from high impact areas.
Matrix Evidence
The dynamic behavior of a Salesforce system goes far beyond the source code. To avoid false-positive, the AI analyzes a multifaceted pool of inputs.
Metadata, Apex & Flows
Structural and versioned truth extracted from Flosum: real source code in production, Validation Rules, metadata of objects and fields, compiled Apex classes and Flows declaratives.
Custom Metadata & Settings
Custom Metadata Types, Custom Settings and parameter data on objects administrative — changing the logic of the code at execution time without deploy.
Event Monitoring & Debug Logs
Productive execution logs and APM strokes allow AI to identify branches of code that are never executed — revealing the zombie code Hidden in the system.
Git & Tacitus Knowledge
Tacit knowledge provides the "Why" Rule exist, complementing the "How" discovered by the machine. Includes history of commits and interviews with experts.
Inventory of Behaviors
We abandon the reductive vision of an "IFs/ELSEs catalog". The final asset is a holistic inventory map all spectrum of systemic behaviors.
Core Business Rules
Logic that translates the commercial operation. "Contracts above R $ 100k require heave Board"
Colateral Effects
Adjacent behaviour (side effects). "Refresh Phase coupling 3 asynchronous processes and 1 call to ERP"
Workarounds & Exceptions
Parametrizated technical debts. "Validation ignored if Profile is SystemAdmin or ID = X"
Integration Matrix
Endpoints consumed and APIs displayed. Inbound/Outbound flow mapping with external dependencies
100% Tier 1 Coverage
All "Critical Level 1" processes with Inventory traced to the line of code.
Bidirectional Traceability
Each behavior linked to a source Unquestionable evidence. Zero rules deduced without ballast.
Human Validation (PR Review)
All artifacts reviewed and approved via Pull Request for a Functional Specialist (SME) and a Tech Lead.
Opacity isolation
Managed packages and opaque integrations tagged as "Black Box" with documented I/O via logs.
Like As-Is Molda the To-Be
The Discovery Baseline feeds each migration decision with exact answers: what to preserve, redesign or discontinue.
Intact Preservation
Components that work properly and have low risk. Migrates without logical change to the new architecture.
Sustained Logic, Tech Changed
Business rule preserved, but modernized implementation. Ex: Workflow Legacy rule → Modern Flow/Appex with scenarios Parity BDD.
Obsolete Code Deleted
Zombie code, historical workarounds and worthless features. Safely removed thanks full traceability of the inventory.
Scenes Given-When-Then extracted become the set Regression officer. If the new architecture passes all the BDD tests in the previous environment, the business parity is guaranteed.
Dependencies Map points to systemic borders ideal for incremental monolith shutdown — without breaking external references Or interrupt operations.
Modelling of Case
From the selection of the critical journey to final deliberation — each stage with governance and traceability.
Critical Journey
Evidence
(MCP)
Markdown
(Fail-Closed)
To-Be (ADR)
Fail-Closed Governance: The model requires a technical leader to validate the inventory via Pull Request. No artifact generated IA continues without explicit human approval — mitigating 99% of the risk of hallucinations leaking for To-Be requirements.
Evaluation of Risks and Sureties
Questioning assumptions and anticipating frictions to design a resilient implementation.
Solid & Proven
- Breakup Discovery vs. Doc Alive: "As-Is" frozen as project with start, middle and end works brilliantly in refactoring.
- Fail-Closed Governance: Review via PR mitiga 99% the risk of hallucinations in To-Be requirements.
- Model Context Protocol (MCP): Local navigation dramatically reduces costs vs. copying entire repository to the LLM context.
It's still Hypothesis
- Complex Apex proficiency: LLMs are spectacular with Python/JS, but Governor Limits, SOQL in Loops and legacy patterns of Trigger can cause initial interpretation errors.
- Causal discovery can confuse local and global impact on Trigger Handlers 20,000+ lines complex.
Open Decisions
- & IP Data Processing: Metadata with IIP hardcoded and sensitive commercial rules require Enterprise/Zero-Retention API layer by Anthropic/OpenAI.
- Processing Cost: Hard caps per pipeline of analysis needs to be defined before execution in scale.
Proof of Concept:
Cancellation in Service Cloud
A single full vertical journey — from selection to inventory approval. When IT and Business see implicit rules and mapped dependencies in days, not weeks, the project will attest unequivocally its value for final modernization.