Generative AI is flattening the layers of Taiwanese companies: strategy moves from “polished at a monthly pace” to “generated at a minute-by-minute pace.” Computable judgment is handed to the system; incomputable choices are left to people. The value of the executive layer shifts from signing off to that of a “chief philosophy officer”: defining the vision and the boundary, building a mechanism of direct connection, and making strategy computable and replayable, so that the organization does not lose its way, or its speed, in a faster beat.
Series: hub | executives (this piece) | middle managers | front line | frontline staff
I|When the “six-minute strategy” becomes real in Taiwan
Generative AI is rewriting the organizational beat of Taiwanese companies across the board. From hardware manufacturing in Banqiao to software startups in Neihu, a five-year blueprint that used to take a team of senior managers months to polish can now, on an AI decision workbench, be chained through “industry benchmarking → sensitivity analysis → multi-scenario stress tests → counterfactual rehearsal,” and in six to ten minutes generate a draft that can be talked with, and computed. Information no longer travels “up the layers and then back down.” It flows back in real time from frontline POS, customer service, cloud ERP, and the data lake, and is fed directly into a conversational decision system. For a CEO or a group vice president, the most immediate feeling is not “I am being replaced,” but “I have to step aside on purpose” — computable judgment is handed to the system; incomputable choices are left to people.
The deeper change is in the sense of time. Local companies are used to “annual planning — quarterly rolling — monthly after-action review.” AI compresses this beat into “continuously generating hypotheses — quickly testing a small sample — reviewing and replaying at high frequency.” The leader’s task shifts from “setting a large map” to “keeping a living map”: calibrating the coordinates at any time, marking the risks, updating the boundary. What the organization needs, then, is not someone who only signs off at the critical nodes, but someone who can steadily confer meaning inside a faster beat.
II|The role remade: from “decision-maker” to “chief philosophy officer”
AI Flattening (AI大扁平化) has not taken power away; it has changed how power is used. AI brings compute to the table, but leaves meaning in the mind. The core value of the executive layer moves to three faces. First, vision and narrative — in a Taiwan where semiconductors, retail, and financial holding companies are all racing at once, leaders have to keep answering “who we are, and what we are to become,” so that the organization still has coordinates it can follow in uncertainty. Second, the value boundary — be clear about which growth not to chase, which data not to touch, which customer groups not to disturb, and hold the long-termism. Third, trust and culture — build the psychological safety of “mistakes are visible, problems can be corrected,” so that experiment becomes a habit rather than a risk.
“Chief philosophy officer” is not a poetic title. It is the ability to turn abstract value into principles that can be executed. Break “customer-first” into “which fields we do not collect by default” and “which groups we do not reach across domains by default.” Break “sustainable governance” into “which margins we will not chase, which supply-chain risks must be calculated first.” When values land as operable boundaries, the thousands of micro-decisions by AI and the front line will accumulate in the same direction.
III|The redistribution of power and information: “conversational leadership” that connects directly to the front line
AI shortens the distance between the war room and the shop floor. Real-time operating boards, conversational analysis assistants, and event-replay engines let executives talk directly with frontline data, and hear a “voice that has not been dressed up.” This is not skipping the chain of command. It is a new mechanism of direct connection: a unified factual base so that everyone looks at the same piece of reality; a consistent language system so that communication is less distorted; a decision trajectory that can be replayed, so that collaboration is more transparent. The exercise of power also shifts from “sign off — send down — report back” to “ask — reach a common view — commit.”
Information abundance, though, enlarges the blind spots. Metrics, model scores, and automatic reports arrive in a flood; executives are easily pulled toward what is “visible,” and miss what has “not yet been measured.” The remedy is to set an “insight discipline” for the organization: for every critical question, define a minimum necessary data set; be clear about the action threshold for when you may act first and when you must observe further; and in meetings, deliberately leave space to discuss “what we still do not know.” Transparency needs order, and order comes from a shared framework for asking questions.
IV|Upgrading the decision workbench: from PowerPoint to “computable strategy”
In most Taiwanese companies, strategy has long been presented in PowerPoint, with the weight on persuasion. In the age of AI, strategy has to be computable, simulatable, and replayable. The first step is to structure the problem: break “e-commerce conversion is falling” into computable sub-questions, define a unified definition and boundary conditions (such as LTV/CAC and payback period), so that AI can reproduce your line of thought, rather than only reciting the conclusion. The second step is scenario rehearsal: attach a “minimum verifiable experiment” to the core assumptions, so that errors are born early and the damage is smaller. The third step is a counterfactual after-action review: after every major decision, leave behind “the worldview and the alternatives at the time,” and later replay and compare, continuously correcting the organization’s collective model.
Go further, and version the strategy. Manage strategy the way you manage code: build a version library for assumptions, the objective function, and the risk list; every adjustment comes with “reason for the change — impact assessment — rollback conditions.” When strategy becomes a system that can be versioned, the speed of learning is enlarged by the system, and no longer depends on word of mouth and memory.
V|Culture and ethics: write values as “constraints that can be executed”
Efficiency accelerates; ethics blurs easily. AI makes growth faster, but “can be done” is not the same as “should be done.” Taiwan has the Personal Data Protection Act, and regulatory requirements from the FSC and the NCC. The chief philosophy officer has to turn values into constraints that can be executed: which groups not to reach, which situations not to use, which data must not be called, how to stop the damage when something goes wrong. A constraint is not a brake. It is a guardrail: so that high-speed travel can be sustained, and so that an agent in the gray zone “knows what it does not know.”
Executable means “accountable to whom, how it is verified, and what happens if it is breached.” So you have to drive a closed loop of values — rules — process — audit: write the rules up front, design the process properly, and run audit as a normal operation. When a deviation happens, do not shame the individual; correct quickly and repair trust. Over the long run, a “value-execution system” recognized by employees and partners is itself a brand asset.
VI|The organizational beat: weave “fast decisions” and “slow principles” together
AI makes action faster, but principles have to become “slow.” Design the beat as weekly: hypothesis — experiment — after-action review; monthly/quarterly: update the hypothesis library and the allocation of resources; half-yearly/yearly: examine whether the principles still apply. Fast is the response to a concrete problem; slow is the guarding of direction and boundary. Only when the two are stacked is the organization resilient.
The beat also includes stop rules and a downgrade process: when a metric breaks a threshold or a risk light is triggered, the system automatically turns to a conservative mode; who holds a temporary veto; how to correct quickly without shaming the person accountable. Once the beat is stable, the team has more safety even while running at high speed — they dare to try, dare to admit a mistake, and dare to iterate.
VII|Talent and the division of labor: let a “human–AI ensemble” compound
Flattening lowers the burden of layers, but it raises the bar for the individual. The key to a manpower rewrite is to reset roles around a human–AI ensemble. Knowledge roles shift from “producing content” to “defining intent and calibrating the output.” Operations roles shift from “chasing problems” to “connecting with exception management.” Management roles shift from “control and approval” to “coaching and enabling.” The logic of promotion has to be updated too: no longer only “how many people you manage, how many tickets you approve,” but “how much other people’s capability you enlarged, how many paths you built, how much reusable knowledge you settled.”
Training becomes learning on the job, cases as the curriculum, AI as the coach: a real task is the course, best practice is the textbook, a conversational assistant is the coach. Only when development and the division of labor are rearranged around a “human–AI ensemble” does the organization jump from “knows how to use the tools” to “has upgraded the structure of capability.”
VIII|Metrics and the board: use “strategy half-life” and “decision lag” to see the truth
The executive board has to move from static targets to dynamic health. Besides revenue, margin, and cash flow, I would bring in three strategic meta-metrics. First, strategy half-life — whether the cycle from a strategy being proposed to needing a major revision is shortening, and whether the shortening is external change or an internal quality problem. Second, decision lag — whether the time from discovery to action is falling, and whether the fall is because authority is clear or because risk control was sacrificed. Third, the rate at which cognitive bias is corrected — the share of major judgments corrected in after-action review, and the time it took; whether the organization is “learning its way toward being more accurate.”
At the same time, bring in path-based after-action review: do not look at the overall average, but along key paths such as “acquisition → conversion → retention” and “order intake → production → shipment,” track follow-up steps, first-pass rate, promise-fulfillment rate, and share actually used. When the board can say clearly “which path got better, why it got better, and whether it can be copied,” management has been promoted from “looking at numbers” to “looking at the mechanism.”
IX|The Taiwan context: find a “minimum viable path” between compliance and landing
Taiwanese companies have to make a careful balance among compliance, security, moving to the cloud, and landing. The Personal Data Protection Act, the FSC’s boundaries for financial cloud and model use, the NCC’s rules on telecom data, and the Labor Standards Act on employee monitoring all have to be written clearly into the top-level design. On cloud strategy, a more workable path is self-building the critical modules plus integrating ecosystem modules: data stored locally (or in a regionalized service that can land), multi-cloud backup, replaceable models, and migratable data/rules, so that you are not locked to a single vendor.
What is more practical is to get one island running first: pick one or two business units or a first-tier city as a pilot, make “strategy — metrics — experiment — financial write-back” into one complete chain, and then copy it across. Externally, building quotable semantic standards and governance norms with local cloud/telecom firms, systems integrators, and industry associations will earn positive feedback faster than “building the whole stack yourselves.”
X|A personal working method: how an executive’s day in Taiwan is rearranged by AI
To put this into practice: in the age of AI, an executive’s day can be arranged like this. In the morning, use a conversational workbench for “three questions at the start of the day” — what are today’s three most important questions? Which three assumptions are the most fragile? Which three small experiments are needed to test them? Before noon, chair a short, dense alignment meeting: around the factual base, not a tug-of-war of opinions; on the spot, turn disagreement into tasks and thresholds. In the afternoon, leave a “quiet period” for 1:1s, attending to the judgment logic and the energy of the people who are accountable, so that you do not talk only about output and never about the person. In the evening, do a counterfactual replay: record the worldview at the time, the options that were abandoned, and the next points to watch.
Fix the rhythm as a leadership SOP: a fixed question list, a fixed hypothesis-card template, a fixed after-action structure. Tools can be replaced; the rhythm has to be kept. Once the rhythm is stable, learning compounds, and the team maintains low friction at high speed.
Closing: make what cannot be replaced into what cannot be wrongly replaced
AI has taken computable work to an extreme, and it has also lit more brightly what cannot be computed. Taiwan’s executives do not have to race the system; they have to ally with time: gather a common view with a clear vision, hold the bottom line with a firm boundary, drive experiment with computable strategy, and accumulate cognitive assets with a trajectory that can be replayed. A “chief philosophy officer” is not a romantic title. It is a practical method of work — so that, in a faster beat, the organization does not lose its way, does not consume itself, and does not lose its speed. In the age of AI Flattening, what is truly scarce is not information or algorithms, but direction, boundary, and trust. Do these three uniquely human capabilities so that they “cannot be wrongly replaced,” and you have completed the upgrade from decision-maker to leader.

David Yang (楊閎) is the founder of Future Insights Lab. He writes AI Future Reflections (《AI 未來反思錄》), with a focus on Agentic Commerce, AIGC, data methods, and organizational renewal.
More deep-dive pieces
- ChatGPT will now show brands how often they are mentioned. That is not proof they were recommended well.
- AI will not only eliminate people: the functions being enhanced, and a manager’s handbook for this week
- AI unemployment ranking: the high-risk functions most likely to be eliminated or largely replaced
- Will AI cause unemployment? Start with the enhance / accelerate / eliminate axes
The judgment, restated
My judgment is that executives will not disappear, but “signing off” will lose its job. The people who remain will have to turn vision into a decision system that can be computed, and replayed.
Three things this week
- Set a rhythm for versioning strategy (who drafts, who sets the boundary, how often it is replayed).
- Run one counterfactual after-action review (if you had not signed off at the time, what would the system have recommended).
- Tighten the decision workbench to a minimum set (benchmarking / sensitivity / stress tests / counterfactuals).
If you are rewriting the executive rhythm, you are welcome to subscribe to Future Insights Lab, and have the next piece in the series arrive in your inbox.
