Ford Ditches Hundreds of AI Cameras and Rehires Retired Engineers

First, let’s look at this fact.

Ford introduced hundreds of AI cameras on its manufacturing lines for tasks like checking design inconsistencies, detecting paint irregularities, and verifying assembly accuracy. This is what is referred to as the “AI-driven visual inspection.” While the investment amount has not been disclosed, industrial AI cameras typically cost thousands to tens of thousands of dollars each. With hundreds of units, it can be estimated that the total investment, including system setup and operation, reached several million dollars.

So, what was the outcome?

Quality declined.

AI cameras can detect what is “visible.” However, the real issues in manufacturing often lie in the “invisible.” Problems that can be overlooked if one is unaware of past design changes, anomalies detectable only by touch or sound, and the vague discomfort of “something feels off” that does not show up in numbers—all of these were outside the scope of the AI cameras.

Ford’s response was clear: they rehired retired skilled engineers, referred to as “graybeards.” They brought back the “eyes” and “intuition” of individuals with decades of experience to the manufacturing line.

What we need to consider here is not that “AI is unusable.” It’s about the design of what was entrusted to AI that was flawed.

OKX Takes the Opposite Approach: A World Where AI Hires AI

On the other hand, the vision put forth by cryptocurrency exchange OKX is entirely different.

OKX envisions an “AI agent economy” where AI agents issue work orders to each other and pay for services in cryptocurrency. This is a world where transactions occur autonomously on the blockchain without human intervention.

Specifically, this is what happens: one AI agent issues a task saying, “I need 1,000 images classified.” Another AI agent accepts the order, processes it, and once completed, payment is automatically made through a smart contract. Humans only need to handle the initial design, and everything else runs on its own.

This is not a far-fetched future scenario. Multi-agent frameworks like AutoGPT and CrewAI are already entering practical stages. What OKX aims to do is layer an “economic layer” on top of that.

While Ford has “removed AI and brought back people,” OKX is saying, “Let’s remove even humans and let AI operate among themselves.” It’s the exact opposite.

So, which approach is correct?

The Breakeven Point is Determined by “Complexity of Judgment × Cost of Failure”

The answer is simple: both are correct, but under different conditions.

To summarize:

Ford’s failure lies in underestimating the complexity of judgment in an area with extremely high costs of failure (automotive quality control) by introducing AI. A single automotive recall can cost hundreds of millions of dollars. They entrusted that judgment to AI cameras, which were unaware of past design history. This was a design flaw.

OKX’s concept can succeed because it operates in areas with low costs of failure and the ability to redo tasks. If an image classification is incorrect, it won’t break a car. In data processing tasks, if something fails, it can simply be re-executed.

SMEs should consider mapping their operations onto this two-axis matrix.

Thinking About the “True Breakeven Point” for SMEs in Numbers

Let’s move beyond abstract discussions and consider concrete numbers.

Let’s assume a local SME (with about 30 employees) is looking to automate its monthly invoice processing.

Current Situation (when done by humans):

• Personnel cost for the accounting staff: 300,000 yen per month (400,000 yen including social insurance)
• Time spent on invoice processing: 40 hours per month (about 25% of total working hours)
• Personnel cost equivalent for invoice processing: 100,000 yen per month

If Automated:

• AI-OCR + automatic bookkeeping tool: 20,000 to 50,000 yen per month
• Initial setup and customization: 100,000 to 300,000 yen (one-time only)
• Processing time: 4 hours per month (including human checks)
• Personnel cost equivalent for checking: 10,000 yen per month

Difference: Cost savings of 50,000 to 70,000 yen per month. Annual savings of 600,000 to 840,000 yen.

This is a project that should be pursued. The cost of failure is low (mistakes in invoice bookkeeping can be corrected). The judgments are clear and rule-based. The data is standardized. It’s a textbook case for AI implementation.

On the other hand, what about automating the “proposal creation for clients” in the same company?

• Generative AI can create drafts. The cost would only be a few thousand yen per month for API fees.
• But what if the quality of the proposals declines and they lose a major contract (worth 5 million yen annually)?
• What are the risks of losing client trust with a “proposal written by AI”?

In this area, the correct approach is to let AI handle the “drafting” and have humans do the “finishing.” It’s not about entrusting everything to AI or having everything done by humans.

Three Things SMEs Should Start Doing Today

The theory is good. So, what should be done in practice?

1. Classify Your Business Operations into Four Quadrants Based on “Cost of Failure × Complexity of Judgment”

There’s no need to identify every single operation. First, separate the operations that directly impact revenue from back-office tasks. Start with standardized back-office operations. This area carries the lowest risk and is likely to yield results.

2. Experiment with AI Tools That Can Be Started for Under 50,000 Yen per Month

Invoice processing, meeting minutes creation, and initial inquiry triaging. These three tasks have tools available that can be tested for just a few thousand to 50,000 yen per month. If no results are seen after three months, you can stop. There’s no need for a multi-million yen initial investment.

3. Clearly Define the “Operations Not to Entrust to AI”

This is the most important point. Ford’s failure stemmed from trying to entrust everything to AI. What is the source of your company’s competitive advantage? If it lies in personal skills or judgment, then that area should not be automated. Instead, automate surrounding tasks to allow those individuals to focus on core operations.

The Lessons from Ford and the Vision from OKX Point to the Same Conclusion

Interestingly, while the stories of Ford and OKX may seem completely opposite at first glance, they actually convey the same message.

“The design of areas to entrust to AI determines everything.”

Ford made a design mistake by introducing AI into quality control, which requires high-level judgment, and failed. OKX is attempting to get the design right by confining AI to clear-rule areas of task ordering, processing, and payment.

The question for SMEs is:

“Which operations in our company can we entrust to AI without incurring significant pain from failure?”

Once that answer is found, you can start automating just that area from tomorrow. You can begin with just 50,000 yen per month. There’s no need to gamble millions of dollars like large corporations.

Rather, the ability to start small and fail small is the greatest weapon for SMEs. Ford needed board approval to remove hundreds of AI cameras. In your company, a single decision from the CEO can bring about change starting tomorrow.

That speed of decision-making is the only structural advantage that allows SMEs to compete with large corporations in the age of AI.