More revenue per person, and a human still in the loop

One idea connects most of what productivity vendors said today: the amount of work a single person can own is going up, and the tools around that person are consolidating and automating to match. The caveat that keeps turning up in the same reports is that a human still has to approve the result.

AI agents are in use, but the sign-off stays human

Start with the distinction that makes this news readable. An automation runs a fixed set of steps you defined in advance: when this happens, do that. An AI agent is given a goal and chooses its own steps to reach it. The second kind is more capable and less predictable, which is exactly why adoption and caution are rising together.

A new survey reports that 88% of enterprises now use or test AI agents, but most still need a human in the loop [1]. Read those two halves together. Wide uptake tells you agents are no longer a pilot-stage curiosity. The second half tells you the hard part has not gone away: someone still reviews what the agent produced before it reaches a customer or a ledger.

For a business choosing tools, the practical test is not whether a product has an agent. It is whether the product makes the review step fast and obvious. A good agent surface shows you what it decided, why, and lets you correct it in seconds. A weaker one hides the decision and asks you to trust it. We wrote earlier about the categories of work that should never be handed off without a person signing the outcome, and that boundary is the one to carry into any agent demo (the decisions automation should never make). The trade-off is real: an agent that asks for confirmation on everything saves you little time, while one that asks for nothing exposes you to its mistakes. The tool worth buying lets you set where that line sits.

Why revenue per employee changes your headcount plan

The second report explains why the agent question matters to your budget. For years, software businesses planned headcount against a simple ratio: for 15 years, B2B ran on one planning ratio that almost no one questioned, where every $200K to $300K of new revenue needed another person [2]. You grew the team roughly in step with revenue. Boards approved plans on that basis and founders raised money to fund the hires.

The same report puts the new median revenue per employee at $237K, a top tier near $655K, and a handful of B2B leaders now clearing $1.5M per head. The number itself is less important than the direction. When a small number of companies run at several times the old ratio, they are not working several times harder. They have changed what one person can cover, and tooling is a large part of how.

For a buyer, this reframes the software decision. The question stops being whether a tool saves a few hours a week and becomes whether it lets one person hold a job that used to take two. That is a higher bar, and it rewards tools that remove whole steps rather than shaving minutes off existing ones. It also argues for consolidation: every tool a person switches between carries a cost in attention that does not show up on the invoice. We have written about why those per-person numbers, not the sticker price, are the figures that should decide a purchase (the numbers that change a decision).

A support-tool merger, and what it asks of buyers

The consolidation theme showed up in a deal. This week Ada, an early player in enterprise AI CX, agreed to merge with Dialpad in an all-share deal valuing the Canadian company at around US$430 million [3]. One party is a focused customer-experience product; the other is a larger contact-centre platform. A point tool is joining a broader suite.

This pattern recurs often enough to plan around. When a specialist product you rely on is absorbed into a platform, you inherit a set of open questions. Will the product keep its own roadmap or bend toward the acquirer's priorities. Will pricing move once it is bundled. Will the integrations you built stay supported. None of these have to go badly, but they stop being under your control.

The trade-off here is the familiar one between a single platform and a set of best-fit point tools. A platform reduces the number of vendors you manage and the seams between them. A point tool gives you depth in one area and independence from a larger company's direction. Neither is correct in general. What a merger like this should prompt is a review: if you depend on a tool that just changed hands, check what you would do if its direction shifts, before you have to. We have covered the case for folding several overlapping tools into one when the overlap is real (replacing five business tools with one), and the discipline is the same from either side.

Automated labelling still wants a human to confirm

Google Workspace returned to the human-in-the-loop idea from a governance angle. Classification labels mark a file's sensitivity, which controls who can open, share, or download it. Increasingly those labels are applied automatically, by rules or by AI reading the content, rather than by a person choosing from a menu.

The update says the company is making it faster to confirm automatically applied classification labels in Google Drive, Docs, Sheets, and Slides [4]. Note the verb: confirm, not apply. The system proposes the label, and a person still accepts it. This is the agent pattern again, in a quieter setting. Automatic labelling scales to the volume of files a real business produces, where manual labelling never could, but a wrong label either exposes a sensitive file or locks up a harmless one. Keeping a confirmation step, and making it quick, is how you get the scale without ceding the judgement. For any tool that touches who can see what, that is the design detail to look for.

A process for getting customers to pay on time

The last item is less about software and more about the cash that pays for it. Late payment is a quiet drain on a small business, and the fix is usually a process rather than a product. The advice today is plain: one operator says they found it fairly easy once they worked out a good process [5]. The components are unglamorous. Communicate early. Notify customers around 60 days before a renewal rather than on the day. Treat the customer with respect while respecting the value of what you sell.

What makes this work is predictability. A customer who hears from you on a schedule, before anything is due, is far more likely to pay on a schedule. The thread back to the rest of the day is direct: the same reliable, ahead-of-time communication that collects payment is what a well-built agent or workflow can carry for you, as long as a person still reads the account before the reminder goes out.

The through-line

Four of today's five items describe the same shift from different seats. More revenue runs through each employee, agents and automated rules take on more of the routine, and vendors consolidate so one person can work in fewer places. In every case the capable version keeps a confirmation step. When you weigh a tool this week, that is the thing to test: not whether it can act on its own, but how cleanly it lets you check it before it does. Yesterday's briefing traced the start of this same pattern (the 8 November briefing).