The day B2B pricing power showed a crack

The thread running through the day is the changing economics of software you already pay for. Vendors are testing how far they can push price and how much routine work an AI agent can absorb, while the ground under both moves as data-access rules tighten around the same tools.

Each item below is worth a few minutes for anyone who buys or renews business software this quarter. Read them together and a pattern shows: the questions to ask a vendor in 2026 are less about features and more about price direction, about what an agent is allowed to do on your behalf, and about who can see what.

A deferred price increase is a signal worth reading

For most of the last four years, business-software pricing moved in one direction. Contracts renewed higher, an AI add-on arrived as a separate line, and buyers paid because switching cost more than grumbling. SaaStr's report frames the run plainly: "For four years straight, the B2B playbook has been the same." [3] The news on the day is that Adobe deferred a big annual price increase, which SaaStr calls the first big crack in that pricing power since 2022. [3]

A single deferral is not a trend, and one vendor's decision tells you nothing certain about your own renewal. But it is the kind of event a buyer should file away, because it changes the balance of a negotiation. When the market assumes prices always rise, a supplier holds the stronger position at renewal. When a large vendor pauses, the assumption weakens for everyone, and the next conversation about a quote is a little more even.

The practical lesson is to treat price direction as a thing you track, not a surprise you absorb. Before you renew, write down what you paid last year, what the new number is, and what actually changed in the product to justify the gap. If an AI feature arrived as a separate charge, decide whether your team uses it enough to pay for it as its own line. A deferral elsewhere is a reminder that the number on the renewal is a proposal, not a fact. We wrote more about reading the offer behind a price in how pricing pages fail and about what you are really buying in what a subscription plan is really selling.

An AI agent working the leads nobody was going to call

The clearest picture of where AI is genuinely useful came not from a keynote but from two operators. SaaStr's write-up of a SaaStr AI 2026 session is direct about that: "It was two operators showing exactly what's working right now in AI agents for sales." [4] The example is specific. An agent was pointed at 8,000 leads a month that no human was going to call, and conversions rose by half. [4]

The number matters less than the shape of the decision. The work handed to the agent was work that otherwise would not have happened at all. Nobody was calling those leads, so nothing was taken away from a person, and the comparison is not human-versus-agent but agent-versus-nothing. That is the strongest case for automation and the one to look for first: a queue of tasks that never gets touched because there is no time to touch it.

This is different from automating work a person does well and cares about. The useful test before you point an agent at anything is to ask what happens today when the work is left undone. If the honest answer is "nothing, because no one gets to it," you have found a good candidate. If the answer is "a person handles it and the outcome depends on judgement," you are in more careful territory. We set out that boundary in how to tell whether a task should be automated, and the harder edge of it — the calls that should stay with a person — in decisions automation should never make.

A second point is quieter but important. An agent working a lead list is only as good as the records it reads. If the contact data is stale, duplicated, or missing the last conversation, the agent inherits every gap. The 50% figure in a demo assumes clean inputs; your own result will track the state of your data. 360REV keeps each customer's history, quotes, and messages on one record so an agent or a person is working from the same picture. That is one sentence of product, and the point stands without it: automation amplifies whatever data you feed it.

Tighter classification for who can see a group's data

While pricing and agents drew attention, Google Workspace shipped a change that speaks to a slower, more structural concern: who inside and outside an organisation can see shared information. Google's note opens with the frame — "Earlier this year, we announced changes to Google Groups to enhance data security and privacy." [1] The rollout brings stricter internal and external classifications for Groups and clearer visual indicators of which is which. [1]

A group or distribution list is one of the quietest ways data leaks. Someone adds an external address to a list that was meant for staff, and months later a thread that assumed a private audience reaches people it never should have. The fix is not dramatic. It is the boring discipline of marking clearly whether a container is internal or external, and showing that mark where people can see it before they hit send.

For a business choosing or configuring tools, the takeaway is to ask how any shared space signals its own boundary. When you set up a mailing list, a shared inbox, or a workspace, the system should make it obvious who is inside the fence. A clear label at the point of use prevents more mistakes than a policy document nobody reads. If you are thinking about the wider question of access as a team grows, why permissions matter as a team grows covers the principle in more depth.

AI arriving inside a tab you already open

The last item is smaller but it illustrates how AI now reaches users. Google is not launching a separate product; it is updating a tab inside a tool people already use. In its own words: "We are introducing several updates to the Gemini tab in Google Classroom designed to make its tools even more helpful for teachers." [2] The stated aim is to help educators work with AI and create visual aids from any device. [2]

The pattern here is worth naming even if you do not run a classroom. AI features increasingly arrive not as a new app you evaluate and adopt, but as a panel that appears inside software you already have. That lowers the barrier to trying them, which is good, and it also means capabilities show up without a decision meeting, which is a thing to watch. When a feature appears in a tab overnight, someone should still ask what data it reads, what it stores, and whether its output needs a human check before it goes out.

The habit to build is a light review whenever a tool adds an AI panel: what does it touch, and who is accountable for what it produces. That is the same question the sales-agent story raises, from the other end. A feature that appears without ceremony still deserves the same scrutiny as one you went out and bought.

What to take from the day

Three of these items point the same way. Pricing power is not fixed, so track your own numbers and treat a renewal as a negotiation. [3] AI agents earn their place fastest on work that would otherwise go undone, and their results follow the quality of the data they read. [4] And the value of any shared or automated system rests on knowing who can see what, which is why a clearer boundary marking is more than housekeeping. [1] The fourth is a reminder that these capabilities now arrive quietly, inside tabs you already open, and deserve a moment's review when they do. [2] None of this is about chasing the newest thing. It is about the small, repeatable questions that keep a stack honest as its price, its automation, and its access all shift at once.