Limits on software that acts on its own
The software news on 19 November keeps returning to one idea: the tools a business runs are starting to take actions on their own, and the harder part is now deciding what they are allowed to touch. The day's releases and reports are less about shiny features and more about controls, benchmarks, and the money moving behind automated software. For anyone picking tools, that shift changes the questions you ask before you buy.
When an automated flow needs a boundary
Start with the concept. An automated flow is a small chain of steps that runs without a person clicking through each one. The useful flows do not just read data; they act. They send a message, update a record, or share a file. The moment a flow can touch a shared resource, you have to think about scope. A flow that helpfully shares a document with a colleague can, if it is pointed at the wrong group, share that document far more widely than anyone intended. That is the quiet risk of automation: the mistake happens at machine speed and across many items at once.
The sensible answer is a boundary that an administrator, not the flow's author, decides. Google Workspace added exactly that. It introduced an admin setting named Collaboration steps that "lets admins control whether Workspace Studio flows can share information with other people or modify shared resources in their organization" [1]. The trade-off is plain. Turn it on and flows can reach across the organisation and do more work for you. Turn it off and you cap the blast radius if a flow is built carelessly. Neither choice is wrong; the point is that a single person sets the limit for everyone, rather than each flow deciding for itself. If you are evaluating any automation feature, ask who holds that switch. We covered the everyday mechanics of automated steps in what automations and macros do for support, and the same lesson applies here: the value is in the control, not just the capability.
What a benchmark tells a buyer
A private software benchmark is a yearly snapshot of how companies in a category are actually performing: how fast they grow, how they price, and what they are spending on. Buyers rarely read these, but they should, because a benchmark tells you what normal looks like. When a vendor quotes you a number, a benchmark is how you judge whether it is ordinary or remarkable.
KeyBanc published the latest edition of its survey, and its weight comes from its history. "KeyBanc Capital Markets just published the 17th edition of its private software benchmark" [2], and this year it added a large section on AI covering pricing, AI costs, and agents. For a business choosing tools, two things matter. First, AI pricing is still being worked out across the industry, so the number you are quoted today may not be the shape of the number a year from now. Second, when a benchmark starts tracking agents and AI costs as a distinct line, it signals that these are no longer experiments buried in a product; they are a cost and a pricing lever in their own right. Read any AI add-on with that in mind, and ask how its price is set and whether it can rise with usage.
Shopping agents reach the checkout
Much of the agent conversation has been about agents that work for a business. The next step is agents that work for a customer, and the holiday season is where that becomes visible. Salesforce published data forecasting a record Cyber Week, with "retailers with shopper agents set to score big as consumers turn to AI and personal agents to find deals this holiday season" [3].
Here is why that matters even if you do not sell online. An agent shopping on a person's behalf does not browse the way a person does. It reads structured information, compares, and decides quickly. If your product data, pricing, and availability are clean and machine-readable, an agent can act on them. If they are inconsistent, the agent may skip you without a human ever seeing the page. The trade-off for a business is effort now against reach later: tidying the data an agent would read is unglamorous work, but it is the cost of being considered by buyers who have handed part of the decision to software.
A marketplace is a signal of fit
When you choose a tool, you are rarely choosing it alone. You are choosing how well it connects to the other tools you already run. A marketplace, the catalogue of apps and integrations that plug into a platform, is a rough measure of that fit. A large catalogue means more of your existing stack can connect without custom work, and the admin controls around it decide who in your team may add what.
Zoom marked a milestone here, describing "4,000 apps, powerful admin controls, and an open platform designed to work the way your business does" [4]. The two parts work together. Breadth is only useful if an administrator can govern it, because every app you connect is another door into your data. The practical lesson for a buyer is to look past the headline count and check whether an administrator can approve, restrict, or review the apps staff install. This is also the quiet argument behind tool consolidation, which we examined in replace five business tools with one: fewer connections mean fewer doors to watch.
The money behind the AI bills
It helps to see the scale of the market a buyer is now part of. SaaStr summarised figures from Anthropic's filing, noting that "Anthropic's revenue went from $41M in 2023 to $386M in 2024 to $4.585B in 2025" [5]. Numbers that size explain the pattern running through the rest of the day. The companies building the models underneath your tools are growing at a pace that pulls AI features into every product category, and that is why pricing, controls, and benchmarks have all moved to the front of the conversation at once.
For a business choosing tools, the through-line is steady. Software that acts on its own is useful, but the value sits in the limits around it: who sets them, what they cost, and how far they reach. The yesterday edition of this briefing, the 18 November roundup, traced the same shift, and today the theme only sharpened.
Sources
- [1] New admin setting to control collaboration steps in Workspace Studio flows — Google Workspace
- [2] KeyBanc’s 2026 AI x Software Survey: 24% Median ARR Growth, Only 6% at the Rule of 40, and 43% of Vendors Watching Customers Vibe Code Their Product — SaaStr
- [3] Salesforce Data: Record Cyber Week Expected to Deliver $75.8B in AI-Driven Sales — Salesforce
- [4] 4,000 apps and counting: What's new in the Zoom Marketplace — Zoom
- [5] The Top 10 Metrics From Anthropic’s IPO for B2B Execs: $33 Billion in Nine Months, 900% NRR, and 12x More $100K Customers in a Year — SaaStr