AI agents move into the admin work of selling and support

A single theme runs through today's announcements: software is moving into the administrative layer that surrounds selling and supporting customers, and the biggest players are paying to own it. Salesforce agreeing to buy an AI customer-service company, a sales-agent report, and a new terminal tool all describe the same shift, and the question underneath every one of them is which parts of a job should be handed to an agent and which should stay with a person.

That question is worth answering before you buy anything, not after. We have written about what AI should and should not do in your business, and today's news is a good test of those principles against real products people are spending money on.

Salesforce agrees to buy Fin for $3.6bn

The headline of the day is a large acquisition. Salesforce has signed an agreement to buy the AI customer-service company Fin. Intercom, which built it, confirmed the deal directly: "We're excited to share that we just signed an agreement for Salesforce to acquire Fin for ~$3.6B." [2] TechCrunch reports the rationale: "Salesforce says it wants to use Fin's team and technology to improve Agentforce, its existing enterprise platform that businesses can use to build custom AI agents that automate tasks." [1]

For a business choosing tools, an acquisition at this scale is not a feature announcement. It is a signal about where a category is heading and about who will own the pieces you depend on. When a large platform buys a capability and folds it into an existing product, two things tend to follow. First, the capability gets easier to reach for existing customers of that platform, because it arrives inside a suite they already pay for. Second, it becomes harder to adopt in isolation, because the vendor's incentive is to sell the suite. Neither of those is good or bad on its own. They are trade-offs you weigh against how much of your operation already sits with that vendor.

The practical lesson is about lock-in. The more of your customer data, your support history and your automation logic lives inside one platform, the more a deal like this decides your roadmap for you. That is not a reason to avoid large platforms. It is a reason to know, on any given Tuesday, what data you could export if the tool changed under you. Ownership of the software can change hands overnight. Ownership of your data should not.

Automating the admin, not the relationship

A report from SaaStr describes the sales version of the same idea, and it opens with a number worth sitting with: "A seller spends 70 to 80 percent of the day not selling." [3] The piece describes agents from a company called Reevo that handle the surrounding work — the account is titled "Automate the Admin, Not the Relationship," and the summary lists what that admin actually is: "Research, meeting prep, note-taking, follow-ups, CRM updates." [3]

The distinction in that title is the whole point, and it is a useful frame for any automation decision. There is work a seller is bad at and resents — the logging, the copying of notes from one system to another, the reminder to follow up in three days. And there is work that only a person can do, which is the actual conversation and the judgement about what a customer needs. An agent that clears the first kind gives time back to the second. An agent aimed at the second kind replaces the thing the customer is paying for.

When you evaluate a sales or support agent, ask which side of that line it sits on. A tool that writes the follow-up draft and updates the record after a call is automating admin. A tool that decides on its own which leads to drop is making a judgement, and you should know it is doing so. We have written about the decisions automation should never make for exactly this reason: the value is in removing the friction around the relationship, not the relationship itself.

The customers you already have

A second SaaStr piece points at a blind spot that no agent fixes for you. Its argument is blunt: "You probably do treat most of your oldest, longest lived customers far worse than a new prospect." [5]

This matters today because every tool above is aimed at acquisition and efficiency — closing more, faster, with less admin. Almost none of it is aimed at the customer who signed years ago and has been quietly paying ever since. That customer rarely gets the attention a new prospect does, because the systems that chase new business are louder than the ones that watch existing accounts.

The fix is not a purchase. It is a habit and a data discipline. You cannot treat long-standing customers well if you cannot see, at a glance, who they are, what they have bought, and when you last spoke. That is a matter of keeping customer data current and of what a customer record should contain. Before buying a tool to win more customers, it is worth asking whether the ones you have would say they are looked after.

AI agents arrive at the command line

GitHub published a beginner's guide to its Copilot CLI, describing it plainly: "Learn how to use slash commands to control your terminal AI agent." [4]

This is a smaller item, but it shows the pattern reaching a new surface. An agent in the terminal is controlled the same way an agent in a chat window is — with typed commands the user has to know exist. The guide is a beginner's overview of those commands, which is itself the point: the barrier to using these tools is no longer capability but discoverability. A tool is only as useful as the commands its user can find. For any team adopting an agent, whether for code or for customer work, the onboarding matters as much as the engine. A powerful agent nobody knows how to steer is shelfware.

A phone line folded into the workspace

Finally, Google Workspace announced Carrier Link for Google Voice. Its description: "Carrier Link allows Workspace customers to easily add phone numbers and calling plans from a certified local carrier, leveraging a pre-configured multi-tenant implementation of SIP Link." [6]

The detail worth noticing is consolidation. This brings a phone number and a calling plan into the workspace a business already runs, rather than standing up a separate telephony system. That is the quiet trend beneath the loud one: not just agents automating tasks, but channels collapsing into fewer places. A phone call, an email and a chat that all land in one system are far easier to act on than three that live apart — the argument we made in what omnichannel means when a customer replies. The trade-off, again, is concentration: the more channels you route through one provider, the more that provider's decisions shape yours. Worth the convenience, usually. Worth knowing you are making the choice.

The through-line for a buyer is steady across all five. Let software take the admin. Keep the relationship, the judgement and the data in your own hands.