AI agents close deals while teams relearn how to focus

The productivity tools publishing today are writing less about new buttons and more about how people and software share the work. The common thread is judgement — deciding which work an AI agent can now carry on its own, which still needs a person, and how to protect human attention while the two work side by side.

For a business choosing tools, that shift matters. A few years ago the question was which product had the longest feature list. The question underneath today's announcements is narrower and more useful: where does automation genuinely take work off a person's plate, and where does it quietly add a new kind of management overhead. If you are weighing systems now, it is worth reading these announcements as evidence about that boundary rather than as product news. Our own view, which runs through how 360REV is built, is that software should do the mechanical work and leave the decisions with the people accountable for them.

AI agents can close deals, but not yet run the whole sale

The most concrete claim of the day comes from SaaStr, which reports running a fleet of AI agents in live revenue work. In its words, "We run 21 AI agents in production at SaaStr" [1], and the company says those agents have booked meetings, revived long-dormant leads, and handled invoicing and collections follow-up. That is a useful data point because it is specific. These are not demos; they are agents doing repetitive, well-defined revenue tasks at a scale and at hours no human team would cover.

The second half of the headline is the part a buyer should weigh most carefully. SaaStr's own framing is that despite all of this, there still isn't a good AI account executive. In other words, the agents are strong at the bounded, rules-driven pieces of a sale — the chasing, the scheduling, the paperwork — and weak at the part that requires reading a person, carrying a relationship, and deciding what to do when the situation does not fit the script.

The practical lesson is to buy automation for the tasks that have a clear definition of done, and to keep a person on the tasks that need judgement. This is the same line we have argued before in what AI should and should not do in your business: automation earns its place on the mechanical work, and loses it the moment a decision carries real consequences for a customer or the business.

Deploying AI is becoming a management discipline

Slack's piece on AI management puts a name to the overhead that a fleet of agents creates. Its stated aim is to teach "what AI management is, why it matters, and how teams can deploy AI responsibly to boost productivity at work" [2]. The key word there is responsibly. Once software is acting on your behalf — writing to customers, moving money, touching records — someone has to own what it does.

That ownership is not free. An AI agent that resurrects a cold lead can also email the wrong person, apply an out-of-date price, or act on a record nobody kept current. Management, in this context, means three things: deciding which decisions an agent is allowed to make, keeping a record of what it did, and keeping the data it reads accurate enough to act on. The second of those is why an audit trail nobody thinks about stops being optional the day an agent starts acting. If software takes an action, you need to be able to see later what it did and why.

This is where the two themes meet. SaaStr shows what agents can do; Slack's framing is a reminder that doing it responsibly is work in its own right. A business choosing tools should ask a vendor not only what the automation can do, but what controls and records sit around it.

Protecting focus without losing collaboration

Alongside the agent stories, Slack published a piece on deep work. Its summary promises to explain "what deep work is, why it matters, and practical ways to improve focus without sacrificing collaboration" [3]. That last clause is the hard part. The same tools that make collaboration easy — the always-on channel, the instant message, the notification — are the tools that fragment attention.

The tension is real and it does not have a clean technical fix. More automation can help by removing low-value interruptions, but it can also add a new stream of agent activity that someone has to watch. For a buyer, the question to ask of any collaboration tool is whether it gives people control over when they are reachable, not only how quickly they can be reached. A product that only optimises for speed of response is optimising for interruption.

The connection to the agent theme is direct. If AI takes the routine chasing and scheduling off people's plates, the value of that is only realised if the time freed up turns into protected, focused work rather than more time spent monitoring the tools.

Flexibility and hybrid work as the setting for all of this

Two further Slack pieces frame the environment the rest of this happens in. One covers workplace flexibility, setting out "what workplace flexibility is, its key benefits, and how flexible work arrangements drive productivity and balance" [4]. The other covers hybrid work, aiming to explain how to build a hybrid workforce "using equity, communication, culture, and productivity strategies that work" [5].

These read as softer than the agent story, but they bear on tool choice more than they first appear. Flexible and hybrid arrangements mean people are not in the same room at the same time, which puts weight on the written record. When a decision is made over a weekend or across time zones, the system has to hold enough context that the next person can pick it up without a meeting. That is a direct argument for keeping customer and project records complete and current — the same argument behind what a customer record should contain. The more distributed the team, the less you can rely on someone simply remembering.

The equity point in the hybrid piece is worth holding onto. A tool that works well for the people in the office but poorly for those who are remote does not create flexibility; it creates two tiers. When comparing products, test them from the position of the least-connected person on the team, not the best-connected.

Communication across generations

Slack's final piece of the day addresses intergenerational communication, offering ways to "bridge intergenerational communication gaps with strategies and tools that help every generation collaborate at work" [6]. The reason this belongs in a tools briefing is that software now mediates most workplace communication, and different people arrive with different defaults — about response times, about formality, about which channel a message belongs in.

A tool cannot resolve those differences, but it can make them worse or better. A system that forces one way of working onto everyone will suit whoever that default matches and frustrate everyone else. One that supports several channels and lets people choose how they engage gives a mixed team room to meet in the middle.

What to take from the day

Read together, the day's announcements describe a workplace where software does more of the doing and people are left with the deciding. AI agents can now carry a surprising amount of bounded, repetitive work [1], but the hard parts of a sale still need a person [1], and running the agents responsibly is a discipline of its own [2]. Around that, the ordinary human questions — how to focus [3], how to work flexibly and fairly [4][5], and how to communicate across a mixed team [6] — have not gone away. If anything, automating the routine makes them more important, because the time and attention it frees up is only worth having if the rest of the system is set up to use it well. When you choose tools, choose for that whole picture, not for the demo.