AI assistants are starting to assign themselves work: should SMEs change how they operate?
Several Silicon Valley companies have recently added "initiative" to their AI assistants, shifting them from question boxes to task entry points. For SMEs, the first thing to change is not the tool but the task you hand it.

AI assistants are moving from passive question boxes to active task entry points. For small and medium-sized enterprises, the first thing to change is not the tool but the task you hand it.
Why are "self-starting" AI assistants suddenly multiplying?
Over the past two months or so, several big Silicon Valley names have queued up to add "initiative" to their AI assistants: xAI's Grok Bot can split work across multiple agents, Meta's Muse helps people handle bills inside WhatsApp, and OpenAI released Dots at its developer conference. According to a roundup by Fengwen author Hualin Wuwang, on 21 October US time, Anthropic offered its own answer, Claude Duet.
One boundary first: these product developments come from Fengwen's roundup, not from each company's official release notes, and the specific feature details cannot be confirmed on the vendors' behalf. What can be confirmed is the trend itself — several companies made product moves in the same direction within a little over a month, which does not look like coincidence.
Past AI assistants shared a common point of fatigue: say nothing and they play dead; say something and they throw every choice back at you. Hualin Wuwang calls this experience "the same kind of fatigue" in the article, which we think is accurate.
Vendors moving toward "proactivity" are essentially solving one problem: AI capability is already sufficient, and the bottleneck is that people do not know how to use it and cannot be bothered to think up prompts every time. Letting AI propose "here are the things that can be done today" lowers the barrier to use.
What does this mean for enterprise users?
It means the standard for evaluating AI tools has to change.
Previously the question was "does it answer well"; now one more question is needed: can it take on a specific, daily, repetitive job? For example, compiling this week's sales data, turning a post-meeting recording into minutes, or sending out a batch of materials on schedule.
There is a trade-off that is easy to overlook. The more proactive the AI, the more clearly you must draw the line in advance between "what can be done automatically" and "what I must confirm". Proactive does not mean letting go, especially for actions such as external publishing and speaking to customers, where one mistake costs far more than the time saved.
The counterargument: isn't this just vendors inventing concepts, with no need for bosses to follow?
That argument is half right.
If a company has not yet identified the thing that goes wrong repeatedly every day, swapping in any AI assistant is a wasted swap. However proactive the tool, it cannot be proactive about a process you have never defined.
But this judgment does not hold under one condition: when you can already state clearly "what must be produced every week, by whom, and where it is sent", then proactivity has value, because what it saves is exactly the time spent re-explaining everything each time.
So not every company should follow. Those whose processes are already stable and who are simply short-handed should.
One thing that can be done now
Do not rush to change tools. First spend half an hour making a list: over the past week, what are the things you and your employees have explained more than three times? Pick one and see whether it can be handed to AI to produce a first draft, with you only making the judgment.
What it cannot do is make decisions for you — after the first draft, the question "is this right?" still has to be answered by the boss.
We agree with something Shu Yi, executive director of Sheyu Jiye, has said: AI writes the first draft; the boss's judgment and real experience are the most valuable part of the content.
As tools become more proactive, human judgment becomes more valuable instead.