Who steers the software and what it remembers

A single question connects today's announcements: how much control does the person using the software keep after the vendor has shipped it. One group of releases is about steering what an AI assistant produces and remembers; the other is about keeping a customer's history and their payments in one place rather than scattered across half a dozen tools.

Customer history that outlives the call

When a customer contacts you a second time, the useful starting point is wherever the first conversation ended. A record that resets between a phone call, a chat, and an email forces the customer to repeat themselves and forces your team to guess. The value of a shared history is not that it is tidy. It is that the next person to pick up the thread can see what was already said, promised, and resolved, without asking the customer to explain it again.

Zoom used its Zoomtopia event to put this idea at the centre of its customer-experience story, framing the takeaway as "customer history that follows every interaction" [1]. The claim worth noticing, for a business choosing tools, is continuity: the record is meant to travel with the customer rather than sit inside whichever channel they happened to use that day. Whether a given product delivers that is a thing to test rather than assume, because the moment a history is split across systems is the moment the continuity breaks. The direction, though, is the one that matters.

This is the plainest version of what people mean when they say omnichannel, and it is worth being precise about the word, which we have taken apart separately in what omnichannel means when a customer replies. In 360REV a contact's calls, messages, and notes sit on one record, so the second conversation does not start from nothing.

Payments that live inside the tool you already use

An integration can mean two very different things. The weaker version sends data from one system to another and hopes both stay in sync. The stronger version puts the second tool's function inside the first, so the user never leaves. The difference matters most for money, because a hand-off between systems is exactly where delays and mismatched records tend to collect, and where a customer notices.

Xero and Bookkeeper360 announced an integration that takes the stronger route. Xero described it as "a first of its kind, using Melio capabilities to power a fully embedded, white-label solution" [2], which means the payment function appears inside the accounting firm's own platform rather than as a separate destination the user is sent to. For a firm, the appeal is obvious: clients stay in one place and the work does not scatter.

The trade-off to weigh is dependence. An embedded, white-label arrangement ties several parties together behind one surface, and a business should know, before it commits, what happens to its payment flow if any one of them changes its terms or its priorities. Integrations that are bolted on rather than built in tend to fail without warning, which is why why your tools do not talk to each other is worth reading before you sign anything.

Giving an assistant standing instructions

Most people use an AI assistant by typing the same context into it over and over. A steadier pattern is to set your preferences and connections once, so the assistant already knows how you want it to respond and which of your tools it is allowed to reach. That turns the assistant from something you instruct afresh each time into something you have configured and can hold to a standard.

Google moved in that direction with a single place to adjust its assistant. It described "a centralized 'customize' hub in the Gemini app, providing a single place to tailor how Gemini responds and connects with your tools" [3]. For a business, the useful part is the second half of that sentence. An assistant that connects to your tools can begin to act on repetitive work rather than only answer questions about it, which is where the time is actually saved.

The caution is the same one that applies to any standing instruction. Be clear about what the assistant may touch, because a configured connection acts without being asked each time, and a convenience set up once can keep acting long after the reason for it has passed. Deciding what belongs to software in the first place is a question we have written about in what AI should and should not do in your business.

Steering output before it is finished

A common frustration with generative tools is that they jump to a finished result you then have to unpick. The alternative is a process you can direct while it runs, so you can correct the shape before the work is done rather than after, when changes are expensive and the thing already looks final.

Gemini Notebook added this kind of control for presentations and video summaries, offering "more creative control over Slide Decks and Video Overviews" [4] through a step-by-step creation flow rather than a single final output. The practical value is that review happens earlier, while the material is still cheap to change. For any team adopting a tool like this, the question to ask is whether the structure it produces matches how your work is actually organised. A tool that lets you steer the draft is only useful if the people using it know what a good result looks like, which is a matter of practice as much as features.

When two assistants start talking to each other

The newest and least settled idea of the day is assistants that coordinate with each other on your behalf. This is capable and it is also where oversight becomes hard, because work can now happen between two systems with no person in the loop at the moment it happens.

SaaStr described this candidly from its own use, writing that "two of our agents now talk to each other every day" and, more pointedly, "we didn't build the connection. They did" [5]. For a business, the lesson is not to avoid this. It is to treat an agent that acts for you the way you would treat an employee who acts with authority: insist on a record of what it did, a clear limit on what it may do, and a way to stop it. The convenience is real, but so is the obligation to keep a trail you can read afterwards.

What to take from the day

The releases point the same way from different angles. Control and memory are becoming the features that separate one tool from another: control over what an assistant produces and what it is connected to, and memory of a customer that survives the channel they happened to use. When you evaluate any of these, test the claim rather than the demo. Ask who stays in charge once the software is doing the work, and ask where the record lives when the conversation moves. The tools that answer both questions plainly are the ones worth a longer look.