Implementation lifecycle

superglue follows the phases an experienced consultant already uses: understand the client, look at the source system, agree the mappings, move the data in controlled batches, reconcile, and keep both systems aligned until go-live is behind you. The agent does the data and configuration work. You make the decisions and approve every write to a client system.

The lifecycle

1. Project context: Connect the places your project knowledge already lives, such as meeting notes, Slack or Teams, email and shared drives. Upload kickoff documents and exports. See Project context.
2. Project setup: Let the agent draft the engagement plan as tasks with owners, dates and dependencies. See Project setup.
3. Client systems: Connect the legacy source, the target ERP and any file exports. Private and on-premise systems connect through the secure gateway. See Client systems.
4. Discovery: The agent inventories entities, volumes, relationships and data quality, and reports readiness. See Discovery.
5. Data mapping: The agent proposes field mappings and transformations and produces reviewable transformed files for the client’s sign-off. See Data mapping.
6. Migration and cutover: The agent loads the target in small approved batches, reconciles source and target, and runs the cutover with the same tools. See Migration and cutover.
7. Post go-live: Scheduled tools keep old and new systems aligned during parallel run and hypercare, with failure alerts. See Post go-live.
8. Handover: Export the mapping, validation and cutover reports, and give the client a complete trace of every run. See Handover.

Not every engagement needs every stage. A data-quality assessment stops after discovery. An integration project skips the migration stage and goes straight to sync tools.

What the agent does and what you approve

The agent does

Reads documentation and live data, drafts plans, proposes mappings, writes and tests the transformation and load tools, reconciles counts, and writes the reports.

You decide

Business meaning of ambiguous fields, what to do with records that cannot map, the migration scope, the cutover sequence, and when a batch is safe to load.

You approve

Every action that writes, updates or deletes data in a client system. The agent shows the target, the mapping and a transformed sample, then waits for your approval.

!

Reads are safe and do not need approval. The agent reads source and target systems freely to explore structure and test mappings. Writes always pause for you.

Deterministic execution

The agent builds the migration as saved tools. A saved tool is a fixed sequence of steps that runs the same way every time, with no AI involved at execution time. That is what makes the migration repeatable: you can run the same load against a test company, then against production, and get the same transformation.

Every run is recorded with its input, each step’s result, and the final output. This record is the audit trail you hand to the client. See Monitoring runs.

Playbooks make it your method

superglue ships with built-in playbooks for the common workstreams: Discover a system, Migrate data, Build an integration and Create an MCP server. Each one encodes the steps above. You can duplicate a built-in playbook and adapt it to your firm’s methodology, templates and sign-off rules, so every consultant runs implementations the same way. See Playbooks.

Where to go next

Start with Project context. If you only want to try the agent on one system first, Discovery is a safe, read-only starting point.