AI stops waiting to be asked, and buyers adjust
A few of Friday's announcements point at the same shift: AI is no longer a feature bolted onto a tool, it is starting to change what the tool does and how you should value it. For a business choosing software this quarter, that changes the questions worth asking — not only what a product does when you open it, but what it does when you do not, and what that costs you in control and in money.
Assistants are starting to act on their own
For most of the past two years, an AI assistant meant a box you typed into. You asked, it answered, and nothing happened until you asked again. Zapier's write-up of personal AI assistants marks the end of that arrangement: "Personal AI assistants have officially stopped being chatbots." [1] The newer tools it lists run in the background, surface the things that matter, and take actions without a prompt each time.
That is a real change in kind, not degree. A chatbot is safe because it is inert between questions. An assistant that runs on its own is useful for the same reason it is risky: it can do things while you are not looking. The practical test when you evaluate one is not how clever its answers are, but how clearly you can see and limit what it is allowed to do unsupervised. Before you hand any tool the authority to act, it is worth being explicit about the calls it must never make without you — we wrote about where that line sits in decisions automation should never make. A 360REV workflow treats an automated action and a human-reviewed one as different settings, so you can grant autonomy in small steps rather than all at once.
The buying question becomes: can I switch the autonomy on for the low-stakes work and keep it off for the rest. If a product only offers all-or-nothing, that is a trade-off to weigh, not a disqualifier.
How software gets priced, and how it gets breached
The commercial side moved the same week. SaaStr's round-up from its 20VC session lists, among other threads, "the API key crime wave, and why seat-based B2B can't get a bid." [2] Two things sit in that short line, and both matter to a buyer.
The first is pricing. Seat-based software charges you per person with a login. That model is under pressure because AI does work that used to need a seat, so counting seats stops tracking the value delivered. For a business, this is not abstract. If a vendor you depend on shifts from per-seat to usage or outcome pricing, your bill can move in either direction, and you want to model that before you are locked in, not after.
The second is the API key. An API key is a long password that lets one program call another on your behalf. When keys leak — committed to a public repository, pasted into the wrong place, left in an old script — whoever finds them can run up charges or reach your data under your own name. As more of your tools talk to each other through keys, the number of these credentials grows, and so does the surface an attacker can use. The defensive habit is dull and effective: rotate keys, scope each one to the narrowest permission it needs, and know which tool holds which. The more autonomous assistants you adopt from the section above, the more keys you are handing out, so the two stories are the same story.
Governance is turning into a setting
Data governance used to be a project with a budget and a committee. It is quietly becoming a checkbox. Google Workspace's weekly recap describes a change to classification labels: "We're making it faster to confirm automatically applied classification labels in Google Drive, Docs, Sheets, and Slides." [3]
A classification label is a tag that marks how sensitive a file is — internal, confidential, and so on — and the rules that follow from that tag, such as who may share it. The shift worth noting is the word *automatically*. The system proposes the label; the human only confirms. That lowers the cost of labelling everything, which is the long-standing reason labelling fails: people will not tag thousands of files by hand.
For a business choosing tools, the lesson generalises beyond one vendor. When you assess a platform, ask whether its safety and governance features are things a person has to remember to do, or things the system does by default and a person merely checks. The second kind survives contact with a busy team. The first kind lapses the first week someone is under pressure.
Reading a claim of leadership
HubSpot announced that it was "Named a Leader in Configure, Price, Quote Solutions," and that its "Revenue Hub receives the highest possible scores in 8 criteria in its first ever evaluation." [4] This is worth using as a worked example of how to read any such claim.
Configure, Price, Quote — CPQ — is the software that turns a salesperson's choices into an accurate, approved quote: pick the options, apply the right price and discount rules, produce a document the customer can sign. It matters because manual quoting is where margin quietly leaks, through the wrong discount or a product combination that was never meant to be sold together.
The useful habit when you see a leadership ranking is to stop at the phrase "the highest possible scores in 8 criteria" and ask which eight. An evaluation measures specific things; a product can top the criteria that were tested and still miss the one that matters to your business. The ranking tells you a serious review happened. It does not tell you the review weighed what you weigh. Read the criteria, then map them to your own deal process before you treat the badge as an answer.
A benchmark for the growth conversation
Finally, a number to calibrate against. SaaStr's look at Palantir notes it "just reported Q3 2026, and it is growing faster at about $9.3 billion in run-rate than it was at $4 billion." [5] Accelerating at that scale is unusual, and the figure is useful precisely because it is an outlier.
When a vendor pitches you on momentum, this is the far end of the distribution. Most tools you will ever buy grow nothing like it, and that is fine — you are choosing software to run your business, not to match a public company's chart. The value of the benchmark is to keep the growth story in proportion. A steady, profitable supplier is often the safer bet than the one promising the steepest curve.
The thread across the day is that AI is changing the product and the contract at the same time. For a careful buyer, the response is the same as it was yesterday: grant autonomy in steps, read the claim behind the badge, and keep your credentials and your labels in order. We tracked the earlier moves in Thursday's briefing, and the direction has not changed.
Sources
- [1] How to automate your AI personal assistant — Zapier
- [2] Jensen Resigns, Sequoia Pulls Back to the Bay, and Anthropic’s IPO Draws 6x Demand: The Top Takeaways from 20VC x SaaStr — SaaStr
- [3] Google Workspace Weekly Recap - November 13, 2026 — Google Workspace
- [4] HubSpot Named a Leader in Configure, Price, Quote Solutions, Q4 2026 evaluation by Independent Research Firm — HubSpot
- [5] 5 Interesting Learnings from Palantir at $9.3 Billion Run-Rate: 96% Growth, 165% NRR, and U.S. Commercial Passing Government for the First Time — SaaStr