OrgGeniusRoadmap · September 2026
Chat with your CRM is now a commodity. Here is what we are building instead.
On 15 September 2026 Anthropic shipped Salesforce in Claude: a plugin with 37 seller skills that reads and writes CRM under the user's permissions. Good. It settles the generic “ask my org a question” layer, and it sharpens what OrgGenius is for. Nothing on this page is shipped unless it says so. Dates are intentions, not promises.
01The thesis
A plugin can talk to your org. It cannot fix it.
OrgGenius is for Sales and Service Cloud orgs that cannot justify Agentforce's price, or are not ready for it. What those orgs need is not another chat window. It is the plumbing underneath: records clean enough that an assistant answers correctly, cold data moved off a metered platform, activity captured as real Salesforce records, documents reachable without a credit meter.
That work has to run inside the org, embedded in the page the person already has open, under their own permissions. A plugin that lives in a chat client is structurally in the wrong place to do it.
So the chat becomes the interface to these modules, not the product. The assistant you have today stays: 90 tools, 13 confirmed actions, cited answers. What changes is what it sits on top of.
What every module shares
- A permission-set-gated feature inside the same package. No second install, no separate product.
- No new infrastructure. Apex, Lightning components and Named Credentials, in your org, and nothing of ours in between.
- Every finding and every archived record carries a plain-language “why” an admin can read without opening Setup.
- Every capability the assistant can reach is registered in the same tool registry it already uses, so it is governed the same way.
02Module A · In build now
Data Readiness Scan
Scan the org for the data and configuration defects that make an AI assistant give wrong answers. Score it, rank it, explain every finding, and hand the admin the query that reproduces it. Runs on install, on demand, and weekly.
Core pack, any org
Twenty checks on the objects every Sales or Service org has: duplicate accounts and contacts, leads that are already contacts, records owned by inactive users, open opportunities with a close date months in the past, cases nobody has touched, picklist values that were retired but are still in use, custom fields nobody ever fills, View All and Modify All spread wider than it should be.
CPQ pack, where CPQ exists
Twenty-six more, switched on only when Salesforce CPQ is installed: duplicate product codes, products with no price book entry, inactive products still inside bundles, price rules and product rules that reference fields that no longer exist, quote templates with merge fields that cannot resolve, approval rules pointing at inactive approvers.
A score you can argue with
Score = 100, minus 15 per Critical check, 6 per High, 2 per Medium. Counted on distinct checks with at least one finding, not on record counts, so one messy object cannot drown the rest. The formula is printed on the dashboard. Muting a check removes it from the score without deleting what it found.
What the assistant does with it
Two tools expose the latest scan to the assistant: the summary and the findings, filtered by category or severity. And when an answer touches a record that has an open Critical finding, the assistant says so in one line, so a wrong number arrives with its own warning label.
Honest status, 15 September 2026: the objects, the base check and the first six CPQ catalog checks are written. The Core pack, the dashboard and the weekly schedule are next. It has not run in a customer org yet.
Every finding carries
A stable check key · a severity · a title · one paragraph on why it matters, meaning the wrong answer or broken quote it causes · the concrete fix · a record count · up to twenty sample records as links · the SOQL that reproduces the full set · a mute toggle
A check that fails is recorded as a finding of its own. The scan never stops because one check threw.
03Module B · Next
Archive to your own storage
Move cold records out of Salesforce and into a bucket or library you already own, at storage cost, with an index row left behind so the record is still findable from the Account page.
One pilot org pays for 100 GB of Salesforce Archive today. The same 100 GB sits in S3 Standard for about $28 a year, or inside an existing Microsoft 365 quota for nothing. That gap is the whole pitch.
An admin writes a policy: which object, which filter, how often, where to. The batch writes each record and its children as one JSON document to the target, writes an index row with up to five searchable fields and a checksum, and only then deletes the original. Storage first, index second, delete third, and the package will not ship a version that does it in any other order.
- Targets: Amazon S3 and SharePoint, through Named Credentials you control
- Archived items listed on the Account, Opportunity or Contact page, readable in place, checksum verified
- Restore, for admins, recreates the record and its children from the document
- Two assistant tools: what is archived for this record, and fetch one
| 100 GB, one year | Cost |
|---|---|
| Salesforce Archive, one pilot org's contract | $9,156 |
| S3 Standard, your own bucket | about $28 |
| SharePoint, inside existing M365 quota | $0 |
Salesforce Backup and Recover is a different product for a different job. We are not replacing it and will tell you to keep it. Status: designed, not yet built.
04Module C · After archive
Activity sync, as real records
Email and calendar activity written into Salesforce as Task and Event records the org actually owns: reportable, shareable, archivable, and visible to the assistant like anything else. Not a side panel that goes dark when the licence lapses.
The readiness scan already looks for the workaround this replaces: staff created as Contacts under personal email addresses so that activity has somewhere to land. Status: specified, not yet built. The first phase needs a Microsoft Graph credential.
05Module D · Last
Document connectors
Reach documents in SharePoint without a credit meter in the path. SharePoint first, because that is where the pilot orgs keep theirs. Google Drive and S3 connectors wait until SharePoint is proven.
Today, files already stored in Salesforce, or surfaced through Files Connect, are searched and read under the person's own file access. This module extends that reach, it does not replace it. Status: on the list, not yet designed.
Why this order: cost to ship and sales impact. The scan needs no external dependency and can be run on any sandbox, so it is also the demo. Archive replaces a line item a buyer can already see on their invoice. Activity sync and connectors each wait on a credential the previous module proves out. Each one ends with something you can install and see, before the next one starts.
This page describes work in progress. The product page describes what is shipped. When the two disagree, the product page is right, and we would rather you asked than assumed.