GPT-6 Astra Integrated into Copilot — A New Era Where 3 Million Yen Outsourcing Estimates No Longer Pass

Let’s get straight to the point.

An AI tool costing 20,000 yen per month has transformed a 3 million yen outsourcing estimate into “too expensive.” This is not a prediction; it is a structural change that is already happening.

OpenAI’s latest model, “GPT-6 Astra,” has been integrated into GitHub Copilot, and general availability has begun. Writing code, reviewing it, and fixing bugs — a significant portion of the work that human developers have traditionally spent time on is now being done by AI “on its own.”

What I want to ask is this: Does your company have any justification for paying the same amount for outsourced development this year as it did last year?

—

What Has Changed — The Cost of Writing Code Has Dropped Significantly

I will leave the technical details of GPT-6 Astra to specialized media. Here, I will only discuss “what changes for small and medium-sized enterprises.”

One thing has changed: The cost of writing code has dropped significantly.

The monthly fee for GitHub Copilot is approximately 2,000 yen for individual plans and about 3,000 yen for business plans. After the integration of GPT-6 Astra, the pricing remains unchanged. Even if you contract the business plan for five users, it amounts to only 15,000 yen per month, or 180,000 yen annually.

On the other hand, when local small and medium-sized enterprises outsource a “slight business system,” the estimates range from 2 million to 5 million yen. With one month for requirement definition, two months for development, and one month for testing, calculating at a man-month rate of 600,000 to 800,000 yen results in 2.4 million to 3.2 million yen. This was the “market rate.”

The premise of this market rate is that “writing code is done by humans, and human time is expensive.” That premise has now collapsed.

I do not claim that an 180,000 yen tool can produce an output equivalent to a 3 million yen outsourcing project. However, the portion of the 3 million yen estimate allocated to “simple coding tasks” — which I estimate to be about 40-60% of the total — has entered a domain that can be replaced by AI. In other words, there is a possibility that work worth 1.2 million to 1.8 million yen could be replaced by a tool costing a few thousand yen per month.

—

Areas Suitable for In-House Development and Areas for Outsourcing — Don’t Misjudge the Boundaries

So, should everything be done in-house? The answer is no. Misjudging these boundaries can actually inflate costs.

Areas suitable for in-house development with GPT-6 Astra + Copilot:

• Building Standard Business Applications: Applications centered around CRUD (Create, Read, Update, Delete) such as customer management, inventory management, and daily report systems. This is already an area where AI excels. If there is someone in-house who is “a bit good at Excel,” they can create something functional in 1-2 weeks by combining it with Copilot. Outsourcing this would cost between 800,000 and 1.5 million yen.

• Modifications and Bug Fixes to Existing Systems: There is no longer a need to wait two weeks and pay 100,000 yen to an outsourcing company for minor fixes like “I want to change this display” or “this calculation logic is incorrect.” By having Copilot read the code and suggest fixes, it can be done in a few hours.

• Rapid Prototyping: When wanting to test a new service idea, there is no need to wait two months for an estimate after outsourcing. A prototype that works with Copilot can be created in three days, allowing internal evaluation. This “speed of trial and error” is the greatest weapon that small and medium-sized enterprises can gain from AI.

Areas that should be outsourced:

• Systems Handling Payments and Personal Information: Security incidents can destroy a company. If you cannot identify security holes in code written by AI, it is best to leave this to experts. However, instead of saying “leave it all to us,” you can show them the AI-generated prototype and ask, “Please just check the security here,” which can significantly reduce outsourcing costs.

• Integration with Core Systems: Connecting with ERP or accounting systems requires specialized knowledge for understanding specifications and testing. The impact of mistakes can be too large.

• Systems Where Performance is Critical: Areas requiring tuning, such as real-time processing of large data sets, still benefit from the experience of human engineers.

The key point is to transition to a hybrid model where you do not go for “everything in-house” or “everything outsourced” but rather “develop in-house with AI and have professionals review just this part.” This alone can reduce outsourcing costs by more than half.

—

The Definition of “One Man-Month” Is Breaking Down — HydraFusion Accelerates Structural Change

Another significant development cannot be overlooked: GitHub’s announcement of “HydraFusion.”

This system allows multiple AI agents to operate in parallel, automatically selecting the optimal model for each task. In other words, one developer can run five AI agents simultaneously, enabling the parallel processing of five tasks.

What this means is the collapse of the unit of “one man-month.”

Traditionally, development companies estimated, “This project will take three man-months.” This could mean one developer working for three months or three developers working for one month. With a man-month rate of 700,000 yen, that amounts to 2.1 million yen. This calculation has been the industry norm.

However, with HydraFusion, it becomes realistic for one developer plus a group of AI agents to complete the equivalent of three man-months of work in just one month. As a result, the client could say to an estimate of “three man-months for 2.1 million yen,” “If we use AI, it can be done in one man-month, right? Please give me a quote for 700,000 yen.”

This is great news for clients, but it poses a critical issue for the service providers — namely, development companies. The premise of the man-month business is collapsing.

For local small and medium-sized enterprises, this presents an opportunity for “the power dynamics with outsourcing partners to reverse.” They can transition from a state of “having to pay whatever is asked because they lack technical knowledge” to a state where they can negotiate with an understanding of market rates thanks to AI.

—

So, What Should We Do?

Here are three actions you can take starting tomorrow.

1. First, contract one license for Copilot and let the person in your company who is best at Excel try it out.
It costs 3,000 yen per month. Even if it fails, it’s not a painful amount. Just experiencing “AI can write code” will change how you view the next outsourcing estimate.

2. When the next outsourcing project arises, first create a prototype with AI before obtaining an estimate.
There is a significant difference in accuracy and cost between verbally communicating “I want something like this” and showing a working prototype. The ambiguity of requirements is the biggest cause of inflated outsourcing costs. By creating a prototype with AI, that ambiguity disappears.

3. Explicitly tell the outsourcing partner, “Please use AI to reduce labor costs.”
If you don’t say it, the outsourcing partner will provide estimates based on traditional man-month calculations. Simply asking for “an estimate based on AI utilization” will change the amount. If it doesn’t change, you should consider changing your outsourcing partner.

—

The Essence of This Change Is the Resolution of “Information Asymmetry”

Why were outsourcing costs high for small and medium-sized enterprises? Because they lacked technical knowledge. When told, “This is a difficult task,” they had no choice but to believe it. Even when looking at the breakdown of estimates, they could not judge their validity.

AI will change this “information asymmetry.” By asking Copilot, “How long will it take to create this feature?” you can get a rough sense of the labor required. If you actually have it create a prototype, you will be able to judge that “200,000 yen is too high for this level of work.”

The democratization of technology means, in essence, “you will no longer be overcharged.”

The structure in which local small and medium-sized enterprises can compete with large corporations lies here. Large corporations take time to change existing outsourcing contracts and internal processes. Small and medium-sized enterprises can change overnight if the owner decides, “We will use this starting tomorrow.” The speed of decision-making can overturn technological disparities.

The era of agonizing over a 3 million yen estimate is over. We have entered an era where you can verify for yourself whether it truly costs that much with a tool costing 3,000 yen per month.

There is no reason not to try.