When agents arrive, pricing and purpose come first

The announcements this week are full of agents that plan, decide and act on their own. What runs underneath nearly all of them, though, is older and plainer: how you are charged for the work, and whether anyone stopped to ask what problem the tool is meant to solve before it was switched on.

Those two questions are worth holding onto, because they are the ones a buyer actually lives with. An agent that drafts a reply or reconciles a ledger is impressive in a demo. The parts that decide whether it was a good purchase are the invoice at the end of the month and the moment, six weeks in, when someone asks what it changed. The reporting from this day lands squarely on both.

The pricing question buyers are already answering for you

G2's 2026 AI Agents Insight Report surveyed 1,019 B2B decision-makers in July 2026 [1]. The headline numbers are about money, not capability: only 9% of buyers think seat pricing is fair, 57% want usage-based pricing, and 74% say they will shop you at renewal. The report's own framing is direct about where to look. As G2 puts it, "The section founders should act on is chapter four, on pricing and renewals" [1].

The reason this matters when you are choosing a tool is that the old per-seat model assumes a human sits in each seat and does a roughly steady amount of work. An agent breaks that assumption. It might do the work of ten seats or sit idle for a fortnight, and a per-seat price charges you the same either way. Usage-based pricing lines the cost up with the work done, which buyers clearly prefer, but it also makes your bill harder to predict and easier to run up by accident. Neither model is free of trade-offs. What the survey tells you is that a large majority of your peers will test the market at renewal rather than accept whatever the first contract said, and that is worth planning for before you sign, not after.

We have written before about reading a plan for what it actually sells you rather than what it advertises, in what a subscription plan is really selling and how pricing pages fail. The G2 figures are a reminder that the pricing model is not a detail at the bottom of the page. For an agent, it is close to the whole decision.

Ask what problem it solves before you deploy

Zoom's latest conversation on customer experience makes the quieter point of the day. The episode opens with an observation most contact-centre leaders will recognise: "Every contact centre leader is under pressure to deploy AI — often before anyone has asked what problem it's actually solving" [2].

That pressure is the thing to guard against. A tool bought to answer a question nobody asked tends to produce activity without result: queues that move faster but resolve nothing, summaries nobody reads, a dashboard that proves the agent is busy. The useful discipline is boring and it works. Name the problem first in plain language, decide how you will know it is fixed, and only then look at which tool fits. An agent is a means, not an outcome. If you cannot state the outcome without naming the tool, you are not ready to buy it yet.

This is the same reasoning we set out in what AI should and should not do in your business and how to tell whether a task should be automated. The question is always the task, never the technology.

The vision being sold alongside the invoice

It helps to see the ambition these products are reaching for, because it explains why the pricing and purpose questions are getting sharper. TechCrunch reports that "Replit CEO Amjad Masad envisions a future of 'self-driving companies,' where AI agents handle day-to-day tasks and people focus on higher-level decisions" [3].

That is a clear statement of direction, and it is worth taking seriously as a direction rather than a description of today. A self-driving company implies agents that carry work end to end with people supervising rather than operating. For a business choosing tools, the practical read is about degree. You do not have to accept or reject the whole vision. You can ask how much of a given task you are willing to hand over now, what you would need to see before handing over more, and what happens on the day the agent does the wrong thing confidently. The vision sets the trajectory. Your own tolerance for unsupervised action sets the pace.

When agents start choosing for you

SaaStr's podcast took the idea a step further into daily operations. In episode #016 of The Agents, the hosts describe a company running "three humans, 21+ agents in production, and a B2B + AI business on track for about $20M this year" [4]. The episode's theme, by its own title, is going "headless" and letting agents pick vendors.

Agents selecting your tools is a genuine shift, and it folds the first two sections together. If an agent is choosing vendors, then the pricing model it meets and the problem definition it works from are no longer things a human reads carefully on a pricing page. They become inputs to a decision you did not personally make. That is not an argument against it. It is an argument for writing down the constraints the agent must respect, the budget it cannot exceed, and the outcome it is buying toward, before you let it loose on a vendor list.

The same shift in the back office

The pattern is not confined to engineering and support. Xero describes bringing a conversational assistant to accounting, opening with the everyday friction it targets: "As an accountant, bookkeeper or small business owner, getting a quick read on your numbers can mean interrupting your workflow" [5]. The promise is a faster read on the figures without leaving what you were doing.

The same caution applies as everywhere else on this day. A quicker answer about cash is useful only if you can trust the number and trace where it came from. Speed on top of data you cannot verify is a liability, not a convenience. Ask the same question you would ask of any agent: what problem does this solve, and how will I know the answer it gives me is right.

For our part, 360REV keeps billing tied to work actually delivered rather than to seats sitting idle, and keeps every automated action traceable back to the record it touched. That is the whole point of the day's thread. The agent is the easy part. The price you pay for it and the problem you bought it to solve are the parts that decide whether it was worth it. For the wider week, our running notes are in the daily briefing for 15 October.