#Data36042 discussing

Hi Alessandro,

Regarding your design pattern for preserving historical audit records from short-lived staging sources in Data Cloud, here are the technical confirmations for how PySpark Custom Code Extensions handle write operations: 

1.WriteMode.APPEND Execution:Append Behavior. 

When using client.write_to_dmo(..., write_mode=WriteMode.APPEND), PySpark strictly appends new incoming rows to the target Data Model Object (DMO) or Data Lake Object (DLO) without truncating or wiping existing data. It will never automatically rebuild or clear out historical rows. 

2.Immunity to Full Refreshes:Platform Independence. 

PySpark code execution is completely separate from the platform-managed Full Refresh schedules used by standard UI Batch Data Transforms. Data Cloud will not automatically trigger a full rebuild or drop table operation on targets populated via Custom Code Extensions. 

3.WriteMode.MERGE and Source Deletions:Merge Behavior on Deletions. 

Under WriteMode.MERGE, records that are deleted or purged from the source staging table remain untouched and preserved in the target table. Merge updates or inserts matching records based on keys, but it does not propagate source deletions to remove historical records from the target.

Hi Pritesh, this one's usually just a deployment/naming issue, not a real bug in the challenge. 

Lead_Home is one of the 6 data streams that gets auto-created when you deploy the Sales data bundle in the Create a Data Stream step (New > Salesforce CRM > View Bundles > Sales > Next > Next > Deploy). If the verifier can't find it, check these in order: 

1. Deployment didn't finish. Deploys can take a few minutes. Go to the Data Streams tab and confirm Lead_Home actually shows up with a completed/active status, not still processing. If it's stuck, cancel and redeploy the Sales bundle. 

2. Wrong path was used. Make sure you picked View Bundles (not View Objects) and selected the Sales bundle specifically — if you added Lead individually through View Objects, Salesforce sometimes won't auto-name it exactly Lead_Home, which breaks the check since it's looking for that exact name. 

3. Deployed twice / renamed it. If you retried the deploy after a failure, you can end up with Lead_Home2 or a duplicate instead of Lead_Home. Delete the duplicate and make sure only one clean Lead_Home exists. 

4. Wrong Data Space. Confirm it deployed to the default Data Space — the challenge checks there, so if you changed the space during setup it won't find it. 

5. Wrong connected source. Double check you selected Salesforce CRM (not Marketing Cloud or another connector) for the source in step 3. 

Fix: delete any bad/duplicate Lead_Home streams, redo New > Salesforce CRM > Sales bundle > Deploy, wait for it to finish (green/active status), then hit Verify Step again. 

Reference:

Hi Venkat, 

Good discussion topic. Rather than picking one winner, here's what the data and Salesforce's own market direction actually show: 

Data 360 is the real foundation, not just another skill on the list. 

Salesforce's own Q3 FY26 earnings confirm Agentforce and Data 360 combined are now their fastest-growing revenue line, up 114% YoY to ~$1.4B ARR, with Data 360 ingesting 32 trillion records in a single quarter (up 119% YoY). More tellingly, Salesforce has architecturally folded Data Cloud into Data 360 as the mandatory data layer underneath Agentforce, agents are only as reliable as the data grounding them. This isn't marketing framing, it's a structural dependency: you cannot build a production-grade agent without clean, governed, unified data underneath it. 

Reference:

Agentforce is the visible layer, Data 360 is the plumbing. 

Most professionals will interact with Agentforce (the builder, agent scripting, orchestration), but the differentiated, harder-to-replicate skill will be Data 360 architecture: semantic data modeling, identity resolution, and governance. Agentforce adoption without solid Data 360 underneath is exactly what's producing "agent quietly creates cleanup work" failures being reported across 2026 implementations. 

Apex/LWC isn't going away, but its role is shifting. 

Programmatic skills remain essential for custom actions, complex orchestration logic, and Apex-invoked Agentforce actions, but they're becoming a supporting skill for AI workflows rather than the primary interface, less "build the whole feature in code," more "build the specific action an agent invokes." 

Integration is quietly becoming inseparable from Data 360. 

With Salesforce's Informatica acquisition now folded into Data 360 (bringing data catalog, MDM, and governance natively), integration skill is merging into the Data 360 skillset rather than remaining a separate specialty. If you're building integration expertise, doing it through a Data 360/MuleSoft lens rather than pure point-to-point API work is the more future-proof framing. 

My honest read: if you can only deepen one area over the next 2-3 years, Data 360 (data modeling, governance, identity resolution) has the strongest structural case, it's the layer every other skill (Agentforce, Apex actions, integrations) increasingly depends on to actually work reliably in production, not just in a demo.