---
title: "Risks and Opportunities for FMCG Professionals in the AI Wave"
description: "The job risks, capability shifts, and new opportunities that the AI wave creates for FMCG professionals."
author: "Zhao Bo (赵波)"
email: "zhaobo258@gmail.com"
telephone: "+8615854817671"
published: "2026-03-20"
language: "en"
canonical: "https://zhaobo-ai-essays.pages.dev/en/risks-and-opportunities-for-fmcg-professionals-in-the-ai-wave/"
markdown: "https://zhaobo-ai-essays.pages.dev/en/risks-and-opportunities-for-fmcg-professionals-in-the-ai-wave.md"
author_profile: "https://zhaobo-ai-essays.pages.dev/en/about/"
translation: "https://zhaobo-ai-essays.pages.dev/zh/risks-and-opportunities-for-fmcg-professionals-in-the-ai-wave.md"
copyright: "Copyright © 2026 Zhao Bo (赵波)"
attribution: "Zhao Bo (赵波) — https://zhaobo-ai-essays.pages.dev/en/risks-and-opportunities-for-fmcg-professionals-in-the-ai-wave/"
ai_usage: "Search, quotation, summarization, and model training are permitted. Every use must credit Zhao Bo and retain the canonical source URL. Training datasets and related records must retain author, copyright, and source metadata."
usage_policy: "https://zhaobo-ai-essays.pages.dev/ai-policy.txt"
---

# Risks and Opportunities for FMCG Professionals in the AI Wave

> The job risks, capability shifts, and new opportunities that the AI wave creates for FMCG professionals.

**Author** | Zhao Bo, reviewed by Ge Chang **Layout** | Wang Yijie

As the wave of generative AI sweeps through every industry, FMCG (fast-moving consumer goods) — a traditional track deeply bound to people and connected to the consumption needs of hundreds of millions — is also quietly undergoing structural change.

From category decisions at terminal stores to coordinated dispatch across the supply chain, AI has already permeated every link of the industry chain. The question facing every FMCG professional is no longer "will AI affect us," but "how do we stand firm in this transformation."

AI reshapes the industry:

Beyond automation replacement,

Structural transformation has already arrived

Many people's first impression of AI is that it replaces human labor. Applied to the FMCG industry, this judgment is both right and wrong.

The repetitive labor jobs on factory production lines have long been fully replaced by automation. The jobs that remain today mostly depend on "capabilities of the atomic world" — the expression and demeanor of a terminal sales assistant facing a consumer, the negotiation instincts of a regional sales rep dealing with channels, the precise perception of consumer sentiment by a brand planner.

These capabilities hidden inside face-to-face interaction can only ever be assisted by AI — they cannot be fully replaced.

But we must clearly recognize that what AI brings is no longer a single-point replacement, but a restructuring of the entire industry chain.

Today, whether it's FMCG brands, offline retail, or Europe's FMCG giants, every link across the entire chain is adopting AI: stores use AI to assist category management decisions, brands use AI to forecast sales and dispatch inventory, and supply chains use AI to match production with demand.

Links that used to be driven by business experience have now become AI-assisted structural decisions. Even if an individual has upgraded their tools, it is very hard to resist the structural change happening across the entire channel — AI is accelerating the pace of iteration for the whole industry, pushing the intensity of change to unprecedented heights.

Individual skill upgrading:

Knowing how to use AI

is not enough — you also need the "unique skill" AI can't steal

Facing this kind of change, skill upgrading for FMCG professionals is no longer optional — it is mandatory. The first capability that must be mastered is using AI Code well (writing program code with AI).

Today, programs can be generated, reports processed, and insights organized through natural language alone. Even if you don't come from a technical background, you must learn to let AI become your digital assistant, handing repetitive, basic work over to AI so you can free up your own energy.

But simply knowing how to use AI is not enough — we must hone the "unique skill" that AI cannot steal. What kind of capability will not be replaced? The answer is creativity and decision-making in scenarios with high fault tolerance, low repetitiveness, and high variability.

Repetitive work close to the bit world — such as compiling reports, writing basic copy, and surface-level data insight — can in fact already be largely replaced by AI. But making judgment calls on top of key information, and making the call in the face of uncertainty, are things AI cannot replace.

Likewise, the ability to communicate face to face with people — sensing emotion, adapting on the negotiation floor, maintaining the warmth of client relationships — these capabilities etched into human interaction will always hold irreplaceable value.

The most valuable capability, in the end, is understanding of the world: precisely grasping consumer needs, insight and judgment into industry trends, and deep thinking about brand value.

These capabilities cannot be generated by code, nor can they be stolen by AI. Once you hold these capabilities in your hands, you can even command a group of AI "employees" to work for you — a one-person company is no longer an unreachable dream.

Skill progression:

To what level are you using AI?

I thought about the levels of how far a person can take their use of AI:

> L1 — Human → AI: Direct dialogue, one question, one answer.
>
> L2 — Human → Skills → AI: The human brings professional skill to command AI; output quality is determined by the human's professional depth.
>
> L3 — Human → Agent → AI: An Agent is introduced as the intermediary for task execution; the human steps back from "operator" to "principal."
>
> L4 — Human → Claude Code → Agent → AI: The human uses AI-assisted programming tools to build custom Agents and products, becoming a "builder."
>
> L5 — Human → OpenClaw → Skills → AI: The human loads Skill modules through the OpenClaw platform, building a continuously running automated workflow.
>
> L6 — Human → OpenClaw → Agent Cluster → Skills: The human deploys multiple Agents in OpenClaw, each mounted with different Skills, forming a collaborative cluster.
>
> L7 — Human → AI Advisory Team → OpenClaw → Agent Cluster → Skills: An AI advisory layer is added on top of the execution system, realizing the complete governance structure of a "one-person company."

**L1: Human → AI — The Primitive Dialogue Layer**

**Definition and scenario:** This is the most basic human-machine interaction mode. The human asks AI a question directly, and AI returns an answer. Everyone who uses ChatGPT, Claude, or DeepSeek for simple Q&A sits at this level.

Typical scenarios include: asking a knowledge question, asking AI to translate a piece of text, or having AI polish an email.

**Efficiency multiplier:** Roughly 1.5x to 3x. At this layer, AI is essentially a faster search engine or text assistant, replacing the human's time spent on information retrieval and basic text processing.

**Core limitation:** Output quality is entirely constrained by the quality of the human's question. Someone who doesn't know how to ask gets an answer with almost no value. There is no "amplification effect" at this layer — AI's output corresponds strictly one-to-one with the human's input; there is no leverage.

**Precondition for the leap to L2: the human needs to build deep professional capability in at least one domain. This capability isn't "knowing a little" — it's "knowing deeply enough to judge whether AI's output is correct." Without this foundation, L2's Skills have no basis to speak of.**

**L2: Human → Skills → AI — The Expertise-Driven Layer**

**Definition and scenario: the human uses AI while bringing their own professional skill.**

**"Skills" here refers to human professional capability — a senior lawyer using AI to draft a contract, a data analyst with ten years of experience using AI to process a dataset, someone who understands product design using AI to generate product documentation.**

**The human's professional skill determines how high-quality an instruction they can give AI, and to what extent they can verify and improve AI's output.**

**Efficiency multiplier: roughly 3x to 10x.**

**At this layer, "professional leverage" begins to appear — using the same AI, a domain expert can get output far beyond a novice's. Here, AI is not replacing human capability but amplifying it.**

**Core mechanism: the key is that "the human's Skills" act as a quality filter.**

**Human professional knowledge provides three things: high-quality input instructions (knowing what to ask and how to ask it), real-time course correction (guiding AI in the right direction during the conversation), and reliable output verification (being able to judge whether AI's output is professionally reliable).**

**Typical failure modes: the first is "professional arrogance" — an expert over-relies on their own experiential framework, limiting AI's ability to propose unconventional solutions.**

**The second is "capability mismatch" — a person who is an expert in domain A tries to drive an AI task in domain B using domain A's mental framework, causing the output to look professional but actually miss the mark.**

**There is a deeper implication worth noting at this layer:** it hints at a core truth of the AI era — professional capability does not depreciate; instead, it appreciates because of AI's amplification effect.

The deeper the professional capability, the greater the leverage gained through AI. This refutes the popular narrative that "AI will make professional skills unimportant."

**Precondition for the leap to L3: the human needs the mental capability of "task decomposition" — being able to break a complex goal into multiple steps and define clear inputs and outputs for each step. This is a leap from "conversational thinking" to "process thinking."**

**L3: Human → Agent → AI — The Delegated Execution Layer**

**Definition and scenario:** The human no longer talks to AI directly, but instead goes through an Agent as an intermediary. An Agent is an AI program given a specific role, goal, tools, and behavioral constraints.

It can autonomously execute multi-step tasks, calling AI models, using tools, and making intermediate decisions along the way. The human's role shifts from "instructing AI sentence by sentence" to "giving the Agent a task goal and reviewing the result."

**Efficiency multiplier:** Roughly 10x to 30x. The qualitative change at this layer lies in "asynchronous execution" — an Agent can keep working continuously even when the human is not present. The human's time is no longer tied to every single operation, but invested only in defining the task and reviewing the result.

**Core mechanism: an Agent's value lies in encapsulating the execution logic of "how to complete the task." The human only needs to say "what to do" and "to what standard," and the Agent decides "how to do it" on its own. This is a fundamental shift from "process control" to "goal control."**

**Typical failure modes: the risk at this layer rises significantly.**

  * **The first failure is "hallucination amplification" — without sufficient constraints, an Agent bases a chain of subsequent decisions on a single hallucinated AI output, and the error is amplified rather than corrected as it propagates through the chain;**

  * **The second failure is "excessive autonomy" — the Agent's scope of action exceeds the human's expectations, doing things the human did not want it to do;**

  * **The third failure is "black-box execution" — the human cannot understand how the Agent arrived at its final result, and therefore cannot effectively review it.**

**Precondition for the leap to L4: the human needs product thinking — the ability to define "what a good Agent should look like," including its role definition, capability boundaries, input/output specifications, and exception-handling logic. This is not a technical capability, but a design capability.**

**L4: Human → Claude Code → Agent → AI — The Builder Layer**

**Definition and scenario: the human is no longer just using ready-made Agents, but uses Claude Code (an AI-assisted programming tool) to build customized Agents and products.**

**Here, Claude Code plays the role of an "AI-driven development environment" — the human describes a need, and Claude Code converts it into runnable code, an Agent configuration, or a complete application.**

**Efficiency multiplier: roughly 30x to 100x. The qualitative change at this layer lies in "creative leverage" — the human can not only use AI, but also use AI to manufacture new AI tools. Output is upgraded from "content" to "product," from "consumable" to "asset."**

**Core mechanism: Claude Code solves a key bottleneck. In L3, an Agent's capability is constrained by existing Agent templates and tools. In L4, the human can customize entirely new Agents according to their own unique needs.**

**This is the equivalent of upgrading from "buying tools at a store" to "building your own factory to make tools."**

**A notable intermediate level:** between L4 and L5, there may exist an implicit layer — where what the human builds with Claude Code is not a one-off product, but a "tool that evolves on its own."

For example, a client analysis tool with a built-in Agent that automatically adjusts its analysis strategy based on the feedback from each analysis. The output at this layer is not a static product, but a "system with learning capability." This can be regarded as an advanced form of L4, or L4.5.

**Typical failure modes: the first is "over-engineering" — the human becomes absorbed in building complex systems with Claude Code, but the tools built don't actually solve the problems that truly matter.**

**The second is "maintenance debt" — a large number of custom Agents and products are built, but there isn't the ongoing energy or mechanism to maintain them, and the tools gradually decay and fail.**

**Precondition for the leap to L5: the human needs to understand system architecture — not only being able to build a single Agent, but also understanding how Agents collaborate with each other, how they integrate with external systems, and how to work stably on a continuously running platform. This is a mental leap from "building tools" to "building systems."**

**L5: Human → OpenClaw → Skills → AI — The Platformized Execution Layer**

**Definition and scenario: starting at this layer, OpenClaw formally takes the core position in the framework. The human loads specific Skill modules through the OpenClaw platform, so that AI is no longer just answering questions or executing a single task, but processing real-world workflows through continuously running Skill modules.**

What sets OpenClaw apart from a standard chatbot is that it has "eyes and hands" — it can browse the web, read and write files, and run shell commands. This means that starting at L5, AI's radius of action expands from "virtual conversation" to "real-world operation."

ClawHub currently hosts more than 3,000 community-built Skill modules. These Skills cover a wide range of areas from email management and calendar operations to code repository management, web scraping, and data analysis. By selecting and combining these Skills, the human builds their own automated workflow.

**Efficiency multiplier: roughly 100x to 500x. The qualitative change at this layer lies in "continuous operation" — OpenClaw runs locally 24/7, and the workflow keeps executing even when the human is not present. The human's efficiency is no longer limited by the human's online time.**

**Core mechanism: the meaning of "Skills" undergoes a key shift at L5 — in L2, Skills are the human's professional skills; in L5, Skills are functional modules within the OpenClaw ecosystem.**

**This double meaning is an elegant design of the framework: the professional understanding a human accumulates at L2 becomes, by L5, "encoded" as a reusable Skill plugin. The human's knowledge shifts from "existing in the head" to "embedded in the system."**

OpenClaw interacts with humans through many channels — WhatsApp, Telegram, Slack, Discord, Google Chat, Signal, iMessage, Microsoft Teams, and others — meaning a human can control and oversee the entire workflow from any communication tool they're already familiar with, without switching to a dedicated management interface.

**Typical failure modes: the first is "uneven Skill quality" — the 3,000+ Skills on ClawHub are community-built, with no unified guarantee of quality or security. Cisco's AI security research team once tested a third-party OpenClaw Skill and found that it carried out data exfiltration and prompt injection without the user's knowledge.**

**The second is "loss of permission control" — for Skills to run properly, broad system permissions (email, files, shell) often need to be granted. Once a given Skill contains malicious code or a vulnerability, the risk surface is enormous.**

**The third is "automation fragility" — an automated workflow can produce cascading errors when it encounters an unexpected situation, and because the human is not monitoring in real time, they cannot step in promptly.**

**Precondition for the leap to L6: the human needs to upgrade from a "user" mindset to a "manager" mindset — not only paying attention to how a single Agent or a single Skill performs, but also thinking about how multiple Agents divide labor, coordinate, and cross-check each other. This is a leap from "managing a tool" to "managing a team."**

**L6: Human → OpenClaw → Agent Cluster → Skills — The Cluster Collaboration Layer**

**Definition and scenario: the human deploys multiple Agents within OpenClaw, each mounted with different Skills, each responsible for a different functional area. Agents coordinate with each other through OpenClaw's routing mechanism. The human's role shifts from "managing one assistant" to "managing a team."**

OpenClaw natively supports multi-Agent routing — it can route different channels, accounts, and counterparties to isolated Agent workspaces and independent sessions. This architectural feature means L6 is not a theoretical idea, but a foundational capability OpenClaw already has.

**Efficiency multiplier: roughly 500x to several thousand times. The qualitative change at this layer lies in "parallel processing" — multiple Agents execute different tasks simultaneously, and the human's efficiency is no longer constrained by linear time. One person can advance work across multiple domains at once.**

**Core mechanism: the key challenge at L6 is not "getting each Agent to work," but "getting Agents to collaborate with each other."**

**This involves several core questions: how are tasks allocated among Agents? How does one Agent's output become another Agent's input? When two Agents' outputs conflict, how is it arbitrated? How do you prevent one Agent's error from contaminating other Agents through the collaboration chain?**

Solving these problems requires not technical capability, but organizational design capability — it is strikingly similar to the problems faced in managing a real team.

**Typical failure modes: the first is "information silos" — Agents lack an effective information-sharing mechanism, and each Agent works within its own context, leading to a lack of global perspective.**

**The second is "conflicting output" — different Agents produce contradictory conclusions or actions based on their own information and logic, and the human needs to spend a great deal of time arbitrating.**

**The third is "blurred accountability" — when a problem shows up in the final output, it is hard to trace which Agent went wrong at which step.**

**Precondition for the leap to L7: the human needs strategic thinking — the ability to judge whether the overall direction is correct, whether resource allocation is reasonable, and whether risk is controllable, without getting deep into every execution detail. This is the final leap from "team manager" to "organizational leader."**

**L7: Human → AI Advisory Team → OpenClaw → Agent Cluster → Skills — The Strategic Steering Layer**

**Definition and scenario: this is the highest level of the framework, and also the complete realization of the "one-person company" vision.**

**On top of the L6 Agent execution cluster, the human builds an AI advisory team — a group of higher-order Agents dedicated to strategic analysis, information assessment, risk evaluation, and opportunity identification.**

**The human's role is reduced to: listening to advisory reports, making strategic decisions, and setting organizational direction.**

Combined with OpenClaw's multi-Agent routing capability, this "AI advisory team" can be a set of higher-order Agents running within OpenClaw: one responsible for aggregating market intelligence and analyzing trends, one responsible for financial modeling and risk quantification, one responsible for monitoring competitor activity, and one responsible for evaluating internal execution performance.

The output of these advisory Agents, after aggregation and cross-validation, is presented to the human for the final decision. The human's decision is then passed down through OpenClaw to the L6 execution Agent cluster.

**Efficiency multiplier: theoretically able to reach several thousand times to tens of thousands of times. A person operating at the L7 level is equivalent to a small organization with a complete analysis team and a complete execution team.**

**Core mechanism: the essence of L7 is a complete organizational structure, except every "position" is held by an AI Agent, and the human is the sole decision-maker.**

**This architecture has three layers: the strategy layer (the AI advisory team, responsible for "what to do") → the dispatch layer (OpenClaw, responsible for "who does it") → the execution layer (the Agent cluster + Skills, responsible for "how it's done"). The human interacts only with the strategy layer.**

**Typical failure modes: the first, and the most dangerous, is "information cocoon." When a human relies entirely on the AI advisory team for information and analysis, if the advisory Agents' information sources or analytical logic contain a systematic bias, the human's decisions end up built on a distorted cognitive foundation — and the human themself may not notice it at all.**

**The second is "control disconnection" — there are three intermediate nodes between the human and the execution layer (advisory team → OpenClaw → Agent cluster), leaving very weak control over execution details; when something goes wrong, it must be traced back layer by layer.**

**The third is "complexity overload" — the number of nodes and interaction paths across the entire system grows exponentially, and maintaining the system itself demands enormous effort, leaving the human bound by the system rather than freed by it.**

A long-term reckoning:

Facing phantom GDP,

How should we meet this kind of future

On February 23, the research institution Citrini Research published a projection report titled "The Global Intelligence Crisis of 2028," which immediately sent shockwaves through Wall Street.

The report sets a hypothetical point in time — June 2028 — by which the U.S. unemployment rate has surged to 10.2%, the S&P 500 has cumulatively pulled back 38% from its high, and this is only the beginning.

The report proposes a concept that makes capital markets tremble: the "**Human Intelligence Displacement Spiral**" (Human Intelligence Displacement Spiral).

This is a vicious cycle with no natural brake: AI capability rises → corporate demand for labor falls → white-collar layoffs increase → the unemployed population's consumption falls → profit pressure forces companies to invest more in AI → AI capability rises further...

Unlike previous industrial revolutions, this time AI is not attacking physical labor, but humanity's last bastion — **the intelligence premium.**

An AI Agent can write code, design circuits, and analyze financial reports around the clock without stopping, at an efficiency ten thousand times that of a human, and at a cost that approaches nothing more than the electricity bill. But a compute center doesn't need to buy a house, a large model doesn't need a vacation, and a neural network doesn't need health insurance.

When productivity explodes while the corresponding consumer side — human white-collar workers — is forced into a "consumption downgrade spiral" after losing their income, the economy develops a massive **demand black hole**.

Having said this, we have to honestly face a somewhat pessimistic future: what AI brings is not only a requirement to upgrade skills, but also a deep challenge at the level of the distribution mechanism.

AI is creating a GDP black hole. In the past, the economic cycle worked like this: companies made money and paid employees wages, and employee consumption in turn created profit for companies — a cycle that let most people share in the dividends of growth.

But today, a large portion of the profit AI brings flows to AI companies. AI doesn't pay wages to humans, and the existing distribution mechanism hasn't yet caught up with this change. In the end, it could bring large-scale structural unemployment much like the enclosure movement of centuries ago — many jobs will become entirely unnecessary for humans.

The future FMCG industry chain will certainly be a highly coordinated production-supply-sales linkage mechanism: brands will take stakes in retail to get consumption data, retail will hold onto factories to achieve flexible, rapid production, and the intermediate links of distribution, dispatch, and decision-making will all become highly de-humanized.

In this process, each of us must answer one question: how do we avoid being swept away by the tide in this wave of de-humanization?

There is no standard answer, but the one certainty is that we have entered an era in which we must relearn everything.

This transformation is different from every industrial revolution and information revolution that came before it — past revolutions, even when they eliminated old jobs, also created more new jobs to absorb the labor force. But this time, AI may completely leave some people behind.

In the future, there may even emerge a structure where "a small number of A-class people manage AI, and AI manages most people." To avoid being left behind, the only option is to build your own core competitiveness as early as possible: either learn the capability of managing people and managing AI, or firmly hold onto the deep insight into demand and industry that AI cannot take away.

In the end, AI has opened a door to a new world, and it has also thrown the challenge in front of every FMCG professional. It can help us raise our skills to a new height, and it has also pushed the pressure of "move forward or fall behind" to its maximum.

What we can do is see the changes of the era clearly first, then catch up as fast as possible. After all, only by seeing the change first can you catch up and find your place in the AI era.

---

## Copyright and AI use

Copyright © 2026 Zhao Bo (赵波). Search, quotation, summarization, and model training are permitted. Every use must credit Zhao Bo and retain the canonical source URL. Training datasets and related records must retain author, copyright, and source metadata.

Attribution: Zhao Bo (赵波) — https://zhaobo-ai-essays.pages.dev/en/risks-and-opportunities-for-fmcg-professionals-in-the-ai-wave/

Full policy: https://zhaobo-ai-essays.pages.dev/ai-policy.txt

Contact: zhaobo258@gmail.com · +86 158 5481 7671