"We want to introduce an AI agent, but we don't know what criteria to use to choose who to build it." As enterprise use of AI agents spreads, we're hearing this kind of question more and more often. (If you'd like to review the basics of AI agents first, see What Is an AI Agent?.)

In its 2025 report "Why AI Projects Fail," RAND Corporation, drawing on interviews with data scientists and engineers, estimates that more than 80% of AI projects end in failure — roughly twice the failure rate of non-AI IT projects. Gartner similarly forecasts that at least 30% of generative AI projects will be abandoned at the proof-of-concept (PoC) stage, citing poor data quality, rising costs, and unclear business value (source: Gartner press release, July 29, 2024).

These figures don't directly represent a "failure rate for choosing a vendor," but many of the failure factors they point to — inadequate understanding of requirements, insufficient data preparation, and so on — are largely preventable through preparation before placing an order and careful vetting of the development company.

This article explains the cost of outsourcing AI agent development, the characteristics of different types of development companies, points to confirm before signing a contract, and common failure patterns.

What to Sort Out Before Outsourcing AI Agent Development

Before you start searching for a development company, there are things you should sort out internally. If this groundwork stays vague, even the best development company will struggle to deliver results.

Clarify "What" and "How Far" You Want to Automate

If you approach a development company with only a vague goal of "we want to introduce an AI agent," the company won't be able to make a precise proposal. It's important to articulate, as concretely as possible, which task, which part of it, how far it should be automated, and where a human should still check the work. Organizing information such as the monthly volume of the target task, the effort it currently takes, and the existing system setup will also improve the accuracy of any quote you receive.

Decide Whether In-House Development or Outsourcing Is More Appropriate

If you have AI and engineering expertise in-house, building the agent yourself using a no-code AI development platform like Dify is also an option. On the other hand, if your business requirements are complex, integration with existing systems is required, or you don't have AI talent in-house, working with an external development company is more realistic. Rather than outsourcing everything, a middle-ground approach — having a company support only the requirements definition while you handle implementation in-house — is also worth considering.

Cost Range for AI Agent Development

The cost of developing an AI agent varies significantly depending on scale and requirements. Based on publicly available cost information from multiple development companies and specialist media, the ranges generally break down as follows.

Scale	Estimated cost	Content
PoC / small-scale verification	¥500,000–¥3,000,000	Verifying feasibility for a specific task; a basic chat UI with limited data reference
Departmental rollout / operational use	¥3,000,000–¥15,000,000	Referencing internal data via RAG, integration with existing systems, permission management
Company-wide rollout / core system integration	¥15,000,000–¥50,000,000+	Rollout across multiple departments, advanced security requirements, integration with multiple systems

On top of this, you'll continue to incur monthly operation and maintenance costs (tens of thousands to several hundred thousand yen) as well as usage-based charges for AI API consumption. AI agents tend to make more API calls per task execution than a typical chatbot, so it's worth confirming the ongoing running costs at the quoting stage, before operations begin.

Cost Breakdown

A typical breakdown allocates roughly 10–20% of the total to the requirements definition and design phase, roughly 40–60% to the development phase (implementing the frontend, backend, AI integration, and external tool integration), and roughly 10–20% to the testing and quality-assurance phase. If you shortchange requirements definition and rush into development, you're likely to run into spec changes and rework downstream, which often ends up inflating the overall cost — so it's important not to cut this phase too short.

The SaaS / No-Code Option

Not every case requires fully custom development. A no-code AI development platform like Dify lets you build an AI agent for a monthly fee starting around a few tens of thousands of yen. For standardized tasks, such a platform is often sufficient on its own — starting at low cost and moving to custom development only once more complex requirements emerge is also an effective approach. A development company that proposes using SaaS where appropriate, rather than defaulting to "let's just build it fully custom," is generally more trustworthy in the long run.

Three Types of Development Companies

Companies that take on AI agent development can be broadly divided into three types based on their business model, each with different strengths and a different type of client they suit best.

Major System Integrators and Consulting Firms

These firms bring a wealth of track record and the capacity to execute large-scale projects, making them well suited to large enterprises and financial institutions with strict security and governance requirements. On the other hand, costs tend to be high, and decision-making and spec changes tend to move more slowly.

AI-Specialized Development Companies

These companies specialize in generative AI, RAG, and AI agents. Compared to major SIers, they tend to be more agile and better suited to a flexible, step-by-step approach that starts from a PoC. Because they focus on this area, you can also expect stronger responsiveness to the latest technology trends. If the company is small, there may be an upper limit on the scale of project it can handle, so it's worth confirming that it's a good fit for your requirements.

Freelancers and Small Teams

These are suited to cases where you want to keep costs down or start with a small-scale PoC. That said, quality can vary significantly, and there is a higher risk if the person handling your project leaves partway through, so it becomes especially important to check their portfolio and track record before signing a contract.

Which type is the best fit depends on whether your requirements are at the "we first want to try something small" stage or the "we want to build this as a company-wide core system" stage.

Seven Points to Check When Choosing a Development Company

1. How Deeply They Listen to Understand Your Business

A good development company doesn't just explain AI technology — it carefully listens to your actual business workflows and the challenges on the ground. A company that goes beyond talking about technology and asks pointed questions like "who handles this task," "at what point," and "what decisions are being made" is more likely to be able to build a system that actually gets used.

2. Whether You Can Confirm Concrete Past Development Track Record

Rather than a vague claim of "extensive experience," confirm, as specifically as possible, what kind of business domains the company has worked in and roughly what scale of project it has handled. Even where confidentiality prevents disclosing details, the company should still be able to describe the industry, business domain, and rough scale involved.

3. Whether They Propose a Phased Approach Starting from a PoC

Rather than jumping straight into full-scale development, whether a company proposes moving forward in stages, starting with a small-scale proof of concept, is an important criterion. That said, a company that doesn't propose a PoC isn't necessarily a bad one — if requirements are already clear, moving directly to full development can also make sense. What matters is whether you can proceed in stages while confirming results along the way.

4. Whether Operations and Maintenance Are Properly Set Up

An AI agent isn't finished once it's built — it requires ongoing operation, including updating reference data, improving response accuracy, and adjusting prompts. Confirm in advance how much post-delivery operation and maintenance the development company will support, and how much of that is included in the monthly fee.

5. Whether the Quote Breakdown Is Clear

Rather than a rough, lump-sum quote, confirm whether the cost of requirements definition, development, testing, and operation/maintenance is broken out separately. A quote with an opaque breakdown carries the risk of additional costs appearing later.

6. How They Handle Data and Security

Since you'll be sharing internal data with an external development company, you need to confirm their policy for handling that data, where it's stored, and how access permissions are managed. If you're dealing with personal or confidential information in particular, check that the contract explicitly spells out terms for handling that data.

7. Whether They Understand Contract Types (Quasi-Delegation vs. Contract for Work)

In AI agent development, a realistic approach is to split contract types by phase: the requirements definition and PoC stage under a "quasi-delegation contract" (paying for effort rather than a deliverable), and full development under a "contract for work" (paying for a completed deliverable). Given that accuracy improvements and spec changes are common in AI development, structuring the entire project as a contract-for-work from the start tends to lead to frequent additional quotes. A development company that can explain the difference between these contract types and why it matters is one that understands the practical realities of AI development.

Common Failures and Red Flags in Quotes and Proposals

There are several red flags worth watching for when choosing a vendor.

A pitch that "AI will solve everything." A company that pushes adoption based purely on technical explanations, without listening to your actual business challenges, carries a higher risk of delivering an "AI nobody uses" after launch.

A quote that's extremely cheap or wildly out of line with market rates. If you're quoted a figure far outside the ranges described above, the scope of work may differ from what you expect, or additional costs may appear later. Always confirm exactly what the quote's scope covers.

No mention of operations or maintenance. If a proposal only covers initial development with no explanation of the post-launch operating structure, you risk being left without support after delivery — a "build it and walk away" outcome.

Jumping straight to large-scale development without proposing a PoC. A company that proposes large-scale development before requirements are settled carries a higher risk of rework and cost overruns down the line.

From Consultation to Contract

A typical process looks like this.

Initial consultation and interview. Many development companies offer the first consultation or interview free of charge. This is where you organize your internal challenges and confirm the general direction of what an AI agent could achieve.

Requirements organization and proposal. Based on the interview, the development company presents a concrete proposal and quote. Get proposals from multiple companies and compare them not just on cost but against the seven points above.

PoC (proof of concept). Before full-scale development, you verify on a small scale whether the AI agent actually functions as intended. This stage confirms whether you're getting the effect you expected.

Full development and rollout. Based on the PoC results, the project moves into full-scale development. It goes through requirements definition, design, implementation, and testing before being rolled out into actual operations.

Operation and improvement. Even after rollout, you continue monitoring usage, updating reference data, and improving response accuracy on an ongoing basis.

Conclusion

The cost of AI agent development ranges widely depending on requirements — from around ¥500,000 at the PoC level to several tens of millions of yen for a large-scale, company-wide rollout. The first step is to clarify exactly "what" and "how far" you want to automate, then consult multiple development companies and compare their quotes.

When choosing a development company, judge based not just on price but on how deeply they understand your business, their willingness to proceed in stages, their operations and maintenance structure, and the transparency of their quotes. Be especially cautious of proposals where introducing AI has become an end in itself, or where there's no explanation of the operating structure.

At Terraverse, drawing on our track record in RAG chatbot development and conversational AI development, we support AI agent development tailored to your company's business requirements. If you'd like to start by simply figuring out which task to tackle first, please feel free to contact us.