Across the past three pieces, we moved from personal efficiency to organizational capability, then posed ten questions to help companies bring the conversation back to the work itself — where, exactly, AI should enter. The short version of where all of that leads: treat AI like a new hire, and most of the confusion clears up on its own. Those ten questions converge on a single idea: strategy has to move top-down, insight has to move bottom-up, and the real design work in human-AI collaboration happens where the two directions actually meet. Where does a top-down strategy begin, and who is responsible for thinking it through? That’s where this piece picks up.
A shallow version of “AI efficiency” is showing up in a lot of companies. An employee who could once explain something in 200 words now expands it into 2,000 with AI, so the material looks more thorough and more professional, then hands it to the boss. The boss, with too much else on their plate to read it closely, feeds the same 2,000 words back into AI and compresses it into a 200-word summary.
AI Adoption’s Center of Gravity
Has Shifted From Tool Use to Redesigning the Work
Over the past two years, most enterprise AI efforts have amounted to a literacy campaign. Companies opened accounts for employees, ran training sessions, built prompt libraries, and encouraged people to use AI for writing, analysis, search and reporting. That step was necessary, but what it mainly improved was individual efficiency. By now, Chinese enterprises are no longer short on AI tools. Feishu, DingTalk, WeCom and a range of AI work platforms have already turned search, writing, analysis, meeting notes and Q&A into everyday features.
As more employees become genuinely capable of using AI in their day-to-day work, the question a company faces changes entirely. It stops being about rolling out AI literacy and becomes: which of these tools are actually helping the company get work done, which tasks no longer need to exist, which tasks should go to AI, which judgment calls still require a person, who reviews what AI produces, and who is accountable when something goes wrong.
If employees are simply using AI to write more reports, produce more proposals and generate more content — without shortening decision cycles, improving the customer experience or growing revenue — then no matter how high the tool-adoption rate climbs, it will rarely add up to enterprise-level productivity.
McKinsey’s enterprise AI research finds that high-performing companies redesign their workflows at roughly three times the rate of average companies. Organizations that see real returns typically aren’t just using more AI — they’re adjusting workflows, management mechanisms and role responsibilities at the same time. Technical capability only converts into financial results once it enters the way the organization actually works.
The direction of national policy has also become clearer. The State Council’s Guiding Opinions on Deeply Implementing the “AI+” Initiative frames AI’s trajectory as a comprehensive shift in industrial development, modes of production and governance capacity. Related measures from the National Development and Reform Commission likewise emphasize the deep integration of AI with the real economy, business scenarios and organizational operations. Translated into what happens inside a single company, the policy direction is actually quite concrete: who does what, who can draw on which resources, who has the authority to decide, and who is accountable for the outcome.
AI is changing the labor structure a company can draw on. What organizations manage in the future may no longer be limited to human employees — it will include robots, digital humans and various kinds of agents. New technology doesn’t benefit every company equally; instead, it rapidly widens the gap between companies in strategic judgment, organizational learning and management capability. Understanding these new productive capacities earlier, and adjusting how the company works accordingly, is quickly becoming a basic capability for staying competitive in this new environment.
Treat AI Like a New Hire, and the Problem Gets a Lot Clearer
When AI comes up, plenty of business owners react the same way: “I don’t understand the technology — this should go to someone who does.” That’s an understandable reaction, but even today it rests on an inaccurate premise: that adopting AI is a technical problem. The more accurate way to see it is that the company has gained a new employee — one who just doesn’t need a desk or a paycheck.
When a company hires a new employee, the manager doesn’t need to fully understand their technical specialty — but they do need to be clear about what the person is responsible for, how much they can decide on their own, who to go to when something comes up, and how performance will be measured. These are, at their core, management judgments — exactly the kind of thing managers already do best. Managing this new employee well takes the same judgment, not a grasp of model architecture or tool operation.
In other words, the real starting point for AI adoption is whether a manager is willing to treat AI like a new hire and take charge of it that way. Learning AI itself isn’t actually the first step. The first version tends to feel intimidating; the second is something managers already do every day — only this time, the person being onboarded is AI.
The Common Mistake in Managing AI
When companies roll out AI, there’s a common default move: hand responsibility to whoever touched the tool first. A Schellman survey found that when asked who is ultimately responsible for enterprise AI, over 40% of respondents pointed to the CIO or IT lead, and only about 10% said it should be the CEO. That default follows the old logic of buying ERP or office software: IT would lead the selection, deployment and rollout, employees would be expected to adapt their workflow once the system was installed, the system’s own capabilities were fixed, and maintenance stayed with IT.
That logic carries real risk in the AI era. AI isn’t a rigid system. Its capability far exceeds any software that came before it, and its boundaries don’t lock in place the moment it’s deployed — it needs to be continuously embedded into day-to-day, end-to-end workflows, and it keeps refining itself as business judgment accumulates.
Which work is worth redesigning, how far AI should go within a process, who’s accountable when something breaks — none of these were ever technical questions to begin with, and IT doesn’t have the authority or the vantage point to answer them. Apply the old logic to AI, and organizational responsibility ends up sitting with whoever has no decision-making power. When problems surface, the organization can only patch individual incidents instead of addressing the root cause.
Which is why the direction, principles and lines of accountability for an organization’s AI use need to be set by the owner and the core leadership team first. Choosing specific tools and working out the technical implementation comes after, and is properly the technical team’s concern. Owners and executives don’t need to master the technical details themselves — what matters is getting the management work right.
Break the Work Apart First, Then Decide Who Does It
Companies traditionally divide labor by role: a marketing manager, a sales manager, a product manager, each holding a full set of responsibilities. But what AI is good at doesn’t map onto an entire role in the organization — it maps onto certain tasks within a role. So when leadership’s first move is to debate “which roles can be replaced by AI,” it tends to manufacture unnecessary anxiety across the organization and misses where the real opportunity actually is.
A better approach starts from the purpose and end-to-end flow of a piece of work. Take a client proposal: it involves gathering information, comparing industries, clarifying requirements, drafting the proposal, making commercial judgment calls, managing the relationship and committing to pricing. AI can take on the information gathering and the first draft; judging client intent, weighing trade-offs and making commercial commitments still needs a person. The outcome of the work hasn’t changed — but how the work gets assembled can be redesigned.
This is also a division of labor managers need to relearn. Work used to be divided between people; now it also has to account for AI, automation systems, robots and digital humans. Managers need to judge which tasks depend on speed and scale, which depend on experience and context, and which involve responsibility that can’t be handed off. Only once these questions are worked through does resource allocation start to make sense.
Leadership in particular has to guard against mistaking “AI can do this” for “this is worth doing.” A weekly report nobody reads doesn’t create value just because generating it now takes three minutes instead of three hours. A company should first decide whether the report is still necessary, which decision it serves and who will act on it — and only then consider who, or what, should produce it.
Owners Must Set the Direction
But Can’t Design Everyone’s Steps
Enterprise AI transformation has to start top-down. Owners and core executives need to be explicit about why the company is using AI, what business outcomes it’s meant to improve, and which principles can’t be compromised. Without agreement on these questions, an eager team below will end up sourcing its own tools and optimizing in isolated pockets — which eventually produces new information silos and organizational friction.
Owners and executives have to make the decisions, but a decision made in a conference room can’t substitute for the understanding that comes from actually doing the work. Leadership is responsible for setting direction, values, risk boundaries and accountability; how a piece of work actually unfolds in practice — what the work involves, how things really happen — usually lives with the frontline employees. They know how to handle a client’s last-minute change of requirements, which fields in the system are never accurate, and how many exceptions hide behind a task that looks perfectly standardized.
The long-term advantage a company builds in the AI era won’t come from any single model or tool. Tools keep updating, and the capability gap between them keeps narrowing. What actually widens the gap between organizations is whether the management team can see the purpose of the work clearly again, cut out internal loops that create no value, and put human judgment and AI capability in the right place relative to each other. Once leadership has set clear goals and a clear accountability framework, there’s a real basis for bringing frontline employees in and letting their real experience shape how the work is designed.
The owner is responsible for deciding why the company is doing this, how far to take it, and which risks are off the table; the team closest to the business needs to answer how the work actually happens, where AI is likely to go wrong, and at what point a person needs to step in. Without both, it’s very hard for a company to move AI from tool use to organizational capability.
Start With One Aligned Leadership Conversation
taasdesign brings Human-Centered Design and design-thinking methods — long applied to commercial, brand and organizational questions — into enterprise AI transformation. We’re opening a small-cohort “AI Work Redesign Workshop” to company owners and core executives, helping leadership teams start from real operating problems, re-examine work goals, task value and the division of labor between people and AI, and build the shared understanding a company needs to move AI forward.
More often than not, what a company is missing isn’t more AI tools — it’s one real opportunity to rethink the work. Tool choices can always be adjusted later; a clear definition of the work and sound management judgment are what actually sustain an organization’s returns.
If your company already has employees using AI but you’re still not sure how to turn individual efficiency into organizational capability — or you’re still working out what it actually means to treat AI like a new hire — we’d welcome the conversation.
