The Work That Took Months is Now ‘Already Finished’ in Just Days

Three months for construction planning. Two weeks for creating a corporate analysis report. A full day for code review.

These tasks are beginning to be compressed into ‘days,’ ‘hours,’ and ‘minutes’ by AI agents. Moreover, instead of waiting for human instructions, the experience of agents autonomously operating and ‘finishing before you know it’ is becoming a reality.

The challenge is not to let this end at ‘amazing.’

Will costs really decrease? From which tasks should we start to recoup our investments? And for small and medium-sized enterprises, is this wave a tailwind or a headwind? Let’s consider this with specific numbers and examples.

—

Construction Planning: From Three Months to Just Days. DeepMind Shows ‘Automation of Judgment’

The case of AI application in construction planning announced by DeepMind indicates that it is not merely about speeding things up. The fact is that AI can replace areas where ‘humans used to make judgments based on experience and intuition.’

Traditionally, formulating construction plans required extensive data collection, site surveys, regulatory checks, and reference to past cases. This is work that skilled professionals would take months to complete. If AI can process this in just a few days, what happens?

First, the structure of labor costs changes. If the monthly compensation for a construction consultant is between 800,000 and 1,500,000 yen, that amounts to 2.4 to 4.5 million yen over three months. If the cost of using an AI agent, including API fees and customization, is between 100,000 and 300,000 yen per month, the cost per project could be just a few hundred thousand yen.

Next, the ‘quality of planning’ is liberated from being dependent on individuals. The accuracy of plans drops when a veteran leaves — a problem that anyone in a local construction company has experienced. Since AI comprehensively references past data, it does not rely on the years of experience of the person in charge. This could become a weapon for small and medium-sized construction companies to bridge the gap with larger firms.

However, there is a caveat. The plans generated by AI are merely ‘drafts,’ and they cannot capture tacit knowledge such as site ground conditions or relationships with neighboring residents. The final judgment must be made by humans. However, 80% of the time spent on ‘information gathering and organization’ before the judgment will disappear.

—

Corporate Analysis: The Reality That ‘Half of AI Insights Are Usable’

The survey results on corporate analysis by agent AI reported by MIT Tech Review need to be taken with a level-headed perspective.

Less than half of the insights generated by AI actually lead to business improvements.

Should we interpret this as ‘AI is still unusable’ or ‘half is usable’? This is the crossroads.

Traditionally, if small and medium-sized enterprises requested corporate analysis from external consultants, they had to be prepared to spend between 500,000 and 2,000,000 yen per project. However, by using AI agents, they can run analyses repeatedly for just a few tens of thousands of yen per month. Even if only half of the insights are usable, the number of trials increases dramatically. If the cost of one consultation allows for 50 analyses with AI, a 50% success rate is more than sufficient.

What is crucial here is to not use ‘AI insights’ as they are, but to create a workflow that uses ‘AI insights as a draft for decision-making in the context of the company.’

Specifically:

1. Input market data, competitive information, and financial data into AI to generate an analysis report (time required: a few minutes to a few hours).
2. Have executives or managers review the insights and compare them with their own sense of the company (time required: 30 minutes to 1 hour).
3. Only adopt those insights that can be turned into actions.

By running this cycle just once a week, over 50 analyses can be conducted annually. This is a stark contrast to relying on external consultants once or twice a year, leading to a vastly different resolution of decision-making.

—

AI Code Review: The Issue of ‘Token Costs Exceeding Labor Costs’

Qodo’s AI code review tool utilizes OpenAI’s Codex, incorporating internal coding standards and past knowledge to detect issues before pull requests. This significantly reduces the time developers spend on reviews.

However, there is a pitfall here.

According to Gartner’s predictions, the token costs incurred when AI performs code reviews may, in some cases, exceed the labor costs of developers. This is especially true when running comprehensive reviews on large codebases, causing API call costs to skyrocket.

Let’s calculate this specifically.

• Monthly cost of one developer (salary + social insurance, etc.): approximately 600,000 to 800,000 yen.
• Percentage of time spent on code review: about 20% (equivalent to approximately 120,000 to 160,000 yen per month).
• Monthly cost of AI code review (including token consumption): a few tens of thousands of yen (depending on the size of the codebase).

For small projects, the cost of AI code review can be kept between 10,000 and 50,000 yen per month. In this case, it costs less than one-third of labor costs, clearly making it pay off.

On the other hand, for large projects where token consumption increases, costs can reach 200,000 to 500,000 yen per month. In such cases, it becomes cheaper to hire a human reviewer, leading to a reversal.

In other words, the break-even point for AI code review is determined by ‘the size of the codebase’ and ‘the frequency of reviews.’ For small and medium-sized development teams (5 to 15 people), there is usually a cost-saving effect. Conversely, this area is one where small and medium-sized enterprises are likely to benefit.

—

What Should Small and Medium-sized Enterprises Automate First? — Three Criteria for Judgment

When told to ‘automate with AI,’ many do not know where to start. This is the reality on the ground.

The criteria for judgment are simply three.

1. Is the Frequency of Repetition High?

Automating tasks that occur only once a month will take time to recoup the investment. Start with tasks that occur several times a week or even daily. Typical examples include processing invoices, initial routing of inquiries, and summarizing daily reports.

2. Is It Dependent on Specific Individuals?

Tasks that can only be done by ‘that person’ are a risk and a cost. By systematizing with AI, business operations will not stop even if the person in charge leaves. The aforementioned formulation of construction plans and the creation of corporate analysis reports fall precisely into this category.

3. Are Current Costs Clear?

You cannot make investment decisions based on ‘it seems difficult.’ Quantify how many hours and how much money are currently spent on that task per month. By placing the monthly cost of AI tools alongside this, the break-even point becomes clear.

Here are some specific examples.

It is clear from this table that the greatest effect of AI automation for small and medium-sized enterprises is in ‘intellectual tasks that were previously outsourced.’ The barriers to in-house production have dramatically lowered due to AI.

—

The Essence of ‘Finishing on Their Own’ is the Collapse of Judgment Costs

The three cases we’ve seen so far — construction planning, corporate analysis, and code review — share a commonality: the costs associated with ‘human judgment’ have drastically decreased.

Collecting data. Organizing it. Comparing it. Referring to past cases. Narrowing down options. These processes are being automated by AI agents. What remains for humans is only the part of ‘deciding whether to do it or not.’

This structural change is more advantageous for small and medium-sized enterprises than for large corporations. Why?

Large companies have specialized departments that handle the processes before judgment. Planning departments, corporate planning offices, quality control departments. If their work is replaced by AI, a massive cost of organizational restructuring arises.

Small and medium-sized enterprises do not have such departments in the first place. The president or site leaders handle ‘researching, thinking, and deciding’ all by themselves. Therefore, if we delegate the researching and thinking parts to AI, the speed of decision-making will increase directly. There is no need for a grand organizational transformation.

—

So, What Should We Do?

1. First, choose one ‘tedious task that you do every week.’ Perfect task selection can wait until later.
2. Quantify how much it currently costs to perform that task. Hourly wage × hours is sufficient.
3. Try it out with free or low-cost AI tools. Starting with ChatGPT, Claude, or NotebookLM is enough. Monthly costs are just a few thousand yen.
4. Evaluate not on ‘whether it can be used or not,’ but on ‘what percentage of the process disappears.’ Do not seek complete automation. If 70% disappears, it’s a huge victory.
5. If it works well, expand to the next task.

The evolution of AI agents will not stop. Waiting will not improve the situation. Try just one thing today. That is the most cost-effective ‘AI strategy’ for small and medium-sized enterprises.