AI Deflation and Vietnam’s Outsourcing Problem

AI deflation does not start with a sudden collapse

In economics, deflation is frightening not simply because prices fall, but because it changes market expectations. When buyers believe the same goods or services will keep getting cheaper, they delay spending, negotiate harder, and force suppliers to deliver more for less. In software, especially outsourcing, a new form of deflation is emerging from the productivity that AI itself creates.

The LinkedIn article “Khi ‘rẻ hơn’ không còn là lợi thế: AI đang viết lại luật chơi của outsourcing” frames this as “AI deflation”. The argument is important because it does not claim that technology jobs will simply disappear. It argues that a large part of technology work is being repriced, especially work that can be described clearly, broken into technical tasks, and accelerated by AI tools.

For Vietnam, the core issue is not whether AI can write code. The deeper issue is the business model many IT services companies have relied on for years: selling software development capacity through developers, man-months, man-days, or hourly rates. When output per developer rises sharply, customers will not treat that productivity gain only as a benefit for the vendor. They will ask why they should still pay for the same number of people.

Low cost used to be a rational strategy

For many years, software outsourcing in Vietnam grew on a clear formula. Customers in the United States, Europe, Japan, Singapore, and Australia needed to scale engineering capacity, but hiring local engineers was expensive. Vietnam offered an attractive alternative: capable engineers, fast learning, strong work ethic, and a cost base significantly lower than mature technology markets.

That formula was not wrong. It created jobs, trained a generation of engineers through international projects, and helped Vietnamese companies enter the global technology value chain. For many customers, timezone differences, communication friction, and cultural distance were acceptable trade-offs when total development cost fell enough.

At that stage, cheaper did not mean worse. Many Vietnamese teams delivered stable output, worked hard, and performed well when requirements were reasonably clear. But the advantage depended on an implicit assumption: offshore engineering was cheap enough to compensate for the added cost of coordination, management, communication, and delivery risk.

AI is weakening that assumption. When a smaller team inside the customer’s own market can use AI to achieve productivity closer to that of a larger offshore team, the cost saving from offshoring becomes less obvious. The costs that used to be tolerated — timezone, culture, security, response time, and business context — become real decision factors again.

Higher productivity does not automatically mean higher value

A common misunderstanding among developers is that AI makes individuals more productive, therefore developers will have stronger bargaining power. At the individual level, this can be true in some contexts. Developers can write tests faster, read codebases faster, generate documentation faster, refactor repetitive logic faster, and explore more implementation options in the same amount of time.

Evidence supports the view that AI can create real productivity gains. GitHub reported that developers using Copilot completed a programming task significantly faster than those who did not use it. McKinsey also found that tasks such as new code generation, refactoring, and technical documentation can be shortened substantially with generative AI, although the gains shrink when tasks are complex or when developers lack familiarity with the framework, domain, or existing system.

But individual productivity does not automatically become pricing power for a service provider. In outsourcing, productivity gains are often converted by customers into demands for fewer people, shorter timelines, or lower prices. If a job previously required ten people for six months, customers will reasonably expect the same job to require five people or three months once the vendor has adopted AI in the delivery process.

That is the uncomfortable logic of AI deflation. AI increases software production capacity while reducing the willingness to pay for work perceived as a commodity. The work still exists, but the market is no longer willing to pay the old price for the same category of work.

Bodyshop models will feel the pressure first

The bodyshop model has a structural weakness in the AI era: revenue is tied more closely to headcount than to business outcomes. When customers buy developers by headcount, the vendor’s value can easily become a procurement calculation. The vendor with enough people, a low enough rate, and an acceptable process has the advantage.

AI distorts that calculation. If one AI-assisted developer can handle the workload previously assigned to several developers, customers have less reason to buy more headcount. The more a vendor proves that AI increases productivity, the more it must explain why the team still needs the same number of people.

This creates a practical dilemma. Without AI, the vendor loses competitiveness because competitors can deliver faster and cheaper. With AI, the vendor undermines the headcount-based billing model that generated its revenue. This does not mean bodyshop outsourcing disappears immediately, but its margins and growth assumptions will face increasing pressure.

That pressure is already visible in India, where large IT services companies are closer to the global outsourcing model at scale. Microsoft announced that Infosys, TCS, and Wipro had each scaled Microsoft 365 Copilot to more than 100,000 employees, exceeding 300,000 seats collectively in less than six months. This is not just an internal tooling story; it shows major technology service providers being forced to turn AI into an operating model.

At the same time, Reuters reported that the chairman of TCS expected AI agents to eventually match employee count and said increased AI usage would slow hiring as tasks become automated. Business Today also reported that the combined market capitalization of India’s five largest IT companies had fallen by more than 46% from its August 2024 record to July 2026, amid weaker demand and GenAI-led pricing pressure.

These figures should not be copied directly onto Vietnam. The market structure, company scale, and customer mix are different. But they are early signals for an industry that shares the same business logic: when software services are sold mainly through the number of people involved, technology that reduces the need for people touches revenue directly.

Fixed-price projects are not immune

Some may assume fixed-price projects are safer because customers buy deliverables, not headcount. In reality, fixed-price work is also affected when AI lowers the production barrier for software. Simple websites, internal dashboards, workflow automation, CRUD applications, and basic integrations are increasingly exposed to pricing pressure.

In the past, a small business might have hired a freelancer or an outsource team to build an internal portal. Today, it can use no-code tools, low-code platforms, AI coding assistants, or one in-house developer who knows how to operate AI to build a first version much faster. That first version may not be production-grade, but it is enough to change expectations about price.

For vendors, the risk is that the easiest parts to demonstrate are often the parts AI produces fastest. UI screens, API endpoints, test skeletons, technical documents, and mock workflows can be generated quickly. Harder concerns such as security models, data governance, legacy integration, observability, migration strategy, performance under load, and production operations are less visible in the early proposal stage.

This creates a difficult paradox. Customers expect prices to fall because they see AI producing visible software artifacts quickly, while vendors still carry the engineering risk that AI cannot own. If a vendor cannot explain architectural depth and delivery risk, it will be pulled into a price war with thinner safety margins.

AI does not kill developers, but it raises the value ladder

Saying that AI replaces developers is too simplistic. DORA 2024 found that AI improves individual productivity, flow, and job satisfaction, but it also brings trade-offs in delivery stability and throughput when organizations lack strong delivery fundamentals. That mirrors real software work: generating code quickly is not the same as delivering reliable software.

High-risk decisions still require humans who can take responsibility. A broken payment system cannot be explained away by saying “AI suggested it”. A multi-tenant architecture that leaks data cannot be blamed on a prompt. A cloud bill that triples due to poor design is not a minor bug. These decisions require understanding domain context, non-functional requirements, data boundaries, security posture, cost models, and long-term operations.

What changes is that the lowest rungs of the value ladder are automated first. A developer who only waits for a clear ticket and implements the spec will be compared with a tool that can generate a fast, cheap draft. A tester who only runs repetitive manual regression checklists faces similar pressure. A business analyst who merely records meetings and turns customer statements into user stories without deeper analysis will also struggle to keep the same value position.

Higher-value professionals are those who can turn ambiguous requirements into clear system models, identify trade-offs, ask the right questions, defend architectural decisions, and use AI as a force multiplier. AI does not devalue this group in the same way. In many cases, it strengthens them because they control the problem, not just the typing speed.

Outsourcing companies need to sell problem-solving capability, not just people

For Vietnamese outsourcing companies, the strategic answer cannot be limited to adding Copilot, Cursor, or a set of AI agents to the delivery workflow and calling it transformation. Tooling is necessary, but it is not a competitive strategy. Once every vendor has AI, AI stops being a differentiator and becomes a cost of entry.

Differentiation must come from problem-solving capability. A higher-value vendor helps customers understand what should be built, why it should be built, where the risks are, and what architecture fits the business context. This requires domain expertise, consulting capability, delivery governance, and the ability to turn technology into outcomes.

FPT is an interesting example of how larger technology service companies are adjusting their narrative. In its 2025 annual report, FPT described digital transformation, AI, and data analytics as important growth areas, with AI and Data Analytics growing 41% year over year. FPT Software also announced an ambition to make AI-First a significant revenue contributor and targeted a 30% productivity increase as part of a three-year roadmap. Each company has different resources and market positions, but the signal is clear: the market is shifting from selling capacity to selling transformation.

Smaller firms do not need to build an AI Factory or a large-scale platform to respond. But they do need to choose domains deeply enough to become differentiated, such as fintech, insurance, education, healthcare, logistics, manufacturing, or enterprise modernization. When a vendor understands the domain, it is no longer merely a task receiver. It becomes a partner that helps customers reduce decision risk.

Solution Architects will stand at the center of this pressure

In the context of AI deflation, the Solution Architect role is no longer just about drawing diagrams or choosing cloud services. Architects must help organizations answer harder questions: what should be automated, what must remain under human control, what should stay in the customer’s core team, what can be outsourced, and what should not be built in the first place.

In traditional outsourcing, architects sometimes appear during proposal or high-level design and then move away from delivery too early. That pattern will become increasingly risky. When AI accelerates implementation, bad decisions also spread faster. A vague design can become a set of wrong modules in hours instead of weeks.

Architects need to be closer to code, closer to business, and closer to delivery risk. Close to code, so they can see whether AI-generated output is creating technical debt. Close to business, so they can separate features that create value from features that only add scope. Close to delivery risk, so they can tell whether AI is truly saving effort or merely shifting effort from coding to review, debugging, and operations.

At the business level, architects also need to help shift the pricing conversation. If a vendor still sells “ten developers for six months”, it will be asked why the same work cannot be done by “five developers in three months”. If the vendor sells “reducing insurance claim cycle time from five days to one day”, or “modernizing a legacy platform while keeping downtime below an agreed threshold”, the conversation changes. When value is measured by business outcomes instead of headcount, AI becomes a delivery accelerator rather than a reason to discount the service.

Developers need to move from writing code to designing small solutions

For individual developers, the practical message is not panic. Panic does not create new capability. But treating AI only as a productivity toy is also insufficient. AI is changing the market’s baseline expectations, and developers need to change accordingly.

A developer in the next phase should learn to read systems, not just tickets. They need to understand why a module exists, which boundaries data crosses, where the failure modes are, and whether the customer is optimizing for speed, cost, compliance, or scalability. With that understanding, developers can use AI to accelerate repetitive work while keeping architectural control.

Prompting is only the surface layer. The more important capabilities are decomposition, review, validation, and judgment. Knowing how to break a large requirement into AI-assisted steps is useful. Knowing how to inspect the output, detect wrong assumptions, ask for alternative designs, and choose the option that fits the context is where real value appears.

Developers should also build portfolios differently. It is no longer enough to list React, .NET, Java, AWS, or Kubernetes. They should show what problem they solved, how cost was optimized, how reliability improved, which workflow was AI-enabled, how much manual operation was reduced, and what technical trade-offs were made under specific constraints. The market does not lack people who can write code. It lacks people who can turn code into reliable outcomes.

A respectful warning is not pessimism

AI deflation is worth taking seriously because it does not rely on the dramatic image of robots replacing humans. It points to a colder economic mechanism: when productivity rises quickly, work without differentiation gets repriced. For an outsourcing industry built heavily on cost advantage, this is a direct pressure.

This is not the end of Vietnam’s IT industry. Vietnam still has a young engineering workforce, fast learning capacity, competitive cost, and international delivery experience. But those strengths must be upgraded into consulting capability, architectural capability, domain expertise, and controlled AI adoption.

The phase where it was enough to be “cheaper but still capable” is narrowing. The next phase will belong to individuals and organizations that can prove “we understand the problem better, design better, deliver more safely, and create clearer outcomes”. AI does not remove all opportunity, but it makes the market less forgiving toward work whose value existed mainly because there was previously no faster tool.

Low cost helped Vietnam’s outsourcing industry enter the global game. But when AI makes software cheaper everywhere, the advantage required to stay in the game will no longer come from price alone. It will come from technical depth, business understanding, accountability, and the ability to create value that customers cannot easily replace with a prompt.

Search

Menu

When Cheaper Is No Longer an Advantage: AI Deflation and the New Problem for Vietnam’s Outsourcing Industry