Agentic Process Transformation
Confirm which execution pattern each application's work should take, before you design any agents. APT reads your IT application inventory and proposes a pattern per system.
Agentic Process Transformation (APT) reads your IT application inventory and, for each application, proposes which agentic execution pattern the work it carries should take. You review each proposal, approve or override it, and the approved set is saved against your session.
The Outcome
Each application in your inventory carries an approved execution pattern, so you know which work is worth building an agent for and which is not.
Before You Start
You should already have chosen the use case you intend to build - see Choosing Your First Use Case. APT is where that choice gets tested against the pattern the underlying work actually needs.
APT works from the documents you uploaded during the AI Assessment. If you did not upload an IT application inventory during the AI Assessment, APT opens on an empty state with an Upload inventory button - add your file directly from that screen.
Start Here
Click Continue to APT in the Jamming Board header to open the APT workspace.
What APT Needs
APT works exclusively from documents you uploaded during the AI Assessment. It reads tables out of spreadsheets (XLSX, CSV, TSV), Word documents, PowerPoint files, and PDFs to find your application inventory. Every proposal it generates is grounded in the rows of your own inventory - nothing is inferred from external sources.
Supported file types: .xlsx, .xlsm, .csv, .tsv, .docx, .pptx, .pdf, .txt, .md, .json
If your inventory is in a PDF without a text layer (a scanned document), APT uses AI to transcribe the tables and flags them as unverifiable - you confirm before analysis runs.
Maximum applications per analysis: 300. If your inventory export is larger (e.g. a full CMDB with servers and components), APT asks you to filter the sheet or accept the first 300 rows.
The Three Execution Patterns
APT proposes one of three patterns for each application:
| Pattern | What It Means |
|---|---|
| Deterministic | Rule-based, structured, and repeatable. Automate with conventional workflow, integration, or RPA. No model in the loop. |
| Agent as Tool | The process path stays fixed, but one bounded step needs judgment a rule cannot express. A model handles that step as a tool inside the flow. |
| Other | Open-ended or sensitive work: retrieval-grounded assistance (RAG), multi-agent orchestration, human-in-the-loop approval, or no agentic opportunity yet. |
How the Analysis Works
APT runs through a phase sequence:
Load uploads
APT reads the documents you uploaded during AI Assessment and lifts tables out of them.
Column mapping
APT identifies which table is your application inventory and suggests which column holds the application name and which holds the business domain (tower). Review and confirm the mapping before analysis runs. If you uploaded a scanned PDF, you confirm the transcribed rows here.
Analysis
APT sends your applications to the model in batches of up to 30 at a time. The model proposes a pattern for each application, cites the evidence from the inventory row that supports the proposal, and flags any cells it could not verify. Your AI Assessment answers (constraints, priorities, and signals) shape which patterns fit - for example, a constraint that regulated reporting must be reproducible shifts pattern selection toward Deterministic.
Review
The review screen shows all proposals organized by business domain (tower) or by pattern. Each application card shows the proposed pattern, confidence level, cited evidence, and rationale. Evidence the model could not verify is struck through.
Approve or override
Approve proposals you agree with. Override any you disagree with by selecting a different pattern and optionally adding a reason. When you are satisfied with the full set, approve the record to save it.
Review Controls
| Control | What It Does |
|---|---|
| By Tower | Groups applications by business domain (the column you mapped as "tower") |
| By Pattern | Groups applications by the three execution patterns |
| All / Proposed / Low | Filter to show all applications, only those not yet decided, or only low-confidence proposals |
| Distribution bar | Shows the count of Deterministic / Agent as Tool / Other across the full set at a glance |
Background Precompute
If you chose the Yes (full AI Readiness) route and uploaded an inventory file during the assessment, APT starts generating proposals in the background while you answer questions. By the time you click Continue to APT from the Jamming Board, proposals are often ready - the workspace loads straight into the review phase.
If no inventory was uploaded during the assessment, APT opens in the upload phase where you can add files directly from the APT workspace.
The Sourcing Guarantee
Every proposal on the APT screen traces back to your inventory rows. The model must quote the cell values it used to reach its conclusion. Quotes it cannot find in the row are flagged automatically. This means you can audit any proposal against your own data without taking the model's word for it.
What Good Looks Like
- The record is approved and saved, not left sitting in the review phase
- The application behind your chosen use case lands on Agent as Tool or Other - a Deterministic proposal means conventional workflow, integration, or RPA is the better answer, and the use case choice should be revisited
- You have used the Low filter and reviewed low-confidence proposals one at a time rather than approving them with the rest
- Struck-through evidence has been checked against the inventory row it came from, or the proposal has been overridden
- Overrides carry a reason, so the decision can be re-read months later by someone who was not in the room
Where This Leads
- Data - the approved patterns tell you what knowledge the work depends on; load it into the platform before agents are designed
- Tool Integrations - an Agent as Tool pattern means an agent has to reach the application that carries the step, so connect that system next
- Governance & Guardrails - the last Align foundation: set the access and compliance boundaries before agents touch real data