If your CRM cannot tell a rep’s update from an agent’s update, every activity report you run next quarter measures something other than your reps.
The numbers behind the problem
On November 3, Salesforce reported that 50% of Slack messages are from apps and agents, and that 90% of Salesforce sellers use Slackbot to update Salesforce. It also described a Vercel inbound sales agent that lets one rep handle lead qualification, research and outreach drafting that would otherwise take a team of 10. These are Salesforce’s own figures about its own product, so I treat them as a direction of travel and not as an audit.
The direction is clear enough to act on. Updates to a deal record are moving into chat, and some share of them are written by software on a rep’s behalf or on its own. A manager who counts calls logged, notes added or fields updated as a sign of effort is already counting a mix of two things.
Advertisement
My position
Every sales team that lets an agent write to its CRM should add an author-type field to the record before it lets that agent run at volume. The field needs three values: a person, an agent acting for a person, and an agent acting alone. Activity reports, pipeline hygiene scores and rep scorecards then filter on it. Without the field, the history cannot be rebuilt later, because the change log will show one user name against edits that came from different places.
The reason is practical. Sales managers use activity as a proxy for effort and as an early warning. A rep whose logged touches fall off gets a conversation. If agents inflate the count for some reps and not others, the warning goes quiet for the wrong people and fires for the right ones in a way nobody can explain. Compensation makes it worse. Any plan that pays on logged activity, even as a small kicker, will reward whoever configured the most helpful agent.
I would also put a number next to every agent: how many records it touched last week, and how many of those edits a person reversed. Reversal rate is the cheapest quality measure a RevOps team has, and it needs the author-type field to compute.
The strongest objection
The best argument against this is that it measures the wrong thing. If an agent drafts the follow-up and a rep approves it, the rep did the work, and splitting the edit into agent and person penalizes the people who adopted the tools fastest. A manager who punishes agent use will get reps who hide it.
I agree with the second half. The field is for visibility, and the scorecard should reward outcomes such as meetings held, pipeline created and deals closed, none of which depend on who typed the note. That is the reason to add the field: it lets a team stop scoring activity at all, because it can show which activity was ever a human signal. The middle value, an agent acting for a person, also protects the adopter, since it records that a rep approved the work.
A second objection is that this belongs to the vendor. Salesforce and its rivals will ship author flags eventually, and a team should wait for them. Waiting costs history. Agents are writing to records now, and a flag added in a later release will not label last quarter’s edits. A custom field takes an admin an afternoon, and a team can map it to the vendor’s flag when one arrives.
What oversight already looks like
Vendors are building controls for the agents themselves. Salesforce’s October 27 Agent Fabric release adds Agent Evals, a kill switch and hard spending limits, and says fewer than half of organizations with agents evaluate them on live traffic. Andrew Comstock, SVP and GM of MuleSoft and Agent Fabric at Salesforce, said: “Enterprises can only scale agents as fast as they can keep them accurate and on budget.”
Accuracy and budget are the two things Comstock names, and a sales team has a third: whose work the record shows. We argued in Name the Owner of Every Sales Agent Before You Write Its Policy that each agent needs a named human owner. The author-type field is the same idea applied to the data. An owner answers for the agent, and the field shows what the agent did.
Spending limits already exist at some vendors, as we reported in Salesforce and HubSpot Now Let Buyers Cap What AI Agents Can Spend. A cap controls cost. It says nothing about whether the numbers a manager reads on Monday came from people.
What I would do this month
Ask the admin to add the three-value field, set it by default for any integration or bot user, and stop using raw activity counts in rep reviews until a full quarter of labeled data exists. Then ask each vendor of an agent that writes to the CRM one question in writing: will every edit it makes carry a flag you can report on? A vendor that cannot answer it is asking you to give up your own sales history.
Source: Salesforce, “Half of Slack Messages Are No Longer From People. Here’s Slack by the Numbers.”
