Outsourcing to India versus Vibe Coding

We’ve been here before. It happened in the late 90s and early 2000s with outsourcing to India: remember the millennium bug, and who were hired to fix it? Yes, Indian outsourcers. The AI rants that dominate LinkedIn feeds today are the same that I used to hear when queuing by the coffee machine (sorry, I don’t do water) twenty years ago.

• “We can’t trust their code.”
• “We’ve seen the code and it’s rubbish; there are bugs, defects, etc.”
• “We are highly regulated and thus, this new way of working doesn’t apply to us”

Indian outsourcers did not win by increasing quality (or not substantially at least in the beginning; given that as they grew their business, they had to expand their workforce which requires an unwavering influx of recent graduates; a.k.a “freshers” in Hinglish). Instead, it was the value proposition which won Western businesses over.

Western business owners who took the plunge when it came to outsourcing to India were perfectly happy to take the hit on product quality provided that costs could be reduced between 25% and 35%. As you can imagine, any product funded by a cost centre would’ve been a prime candidate for outsourcing. We have to remember that in many businesses, especially legacy ones, the entire IT department is classified as a cost centre.

Businesses only made an exception for products in which revenue generation was directly linked to product quality. Eventually, Indian outsourcers caught up with this pattern and ended up acquiring boutique consultancies whose sole unique selling proposition (USP) was ’local talent’ and ‘quality’ outcomes. I’ve been on both sides of this equation. Whether businesses wanted quality, or scale (or maybe both), an India-based hegemon would always be ready to answer the call.

Indian outsourcers, in partnership with clients, also cracked the compliance dilemma. From structured blended teams (consisting of cleared members deployed to the clients’ relevant jurisdictions), to pixel-streaming technology (using remote desktop services so that all storage is local to the client’s jurisdiction), to auditable, military-grade physical facility security, a way for Western businesses to work effectively with Indian providers would eventually be devised.

Today, notwithstanding the rise of the service providers based in the Philippines and Eastern Europe, India remains the world’s IT services hub. Needless to say, a significant portion of the delivery effort in these emerging markets is carried out by subsidiaries of Indian hegemons or Western outsourcers (e.g., Accenture) whose core delivery capability is by and large based out of India.

Having helped businesses work effectively with outsourcing partners (both from a client and service-provider perspective), I see no chance whatsoever of any business adopting AI (when it comes to product development, aka “Vibe Coding”) with the same caution or kid gloves approach I witnessed during the transition to India-based outsourcing. In fact, I see stark differences in how development was outsourced to India and the current AI Adoption trend, at least as far as Vibe Coding is concerned:

What I illustrate above is my summary of how business leaders perceive AI value rather than what effective and proven industry results might be. I’ll elaborate further on each point:

Value Proposition

Cost

For business leaders, AI cost is not theoretical. They are paying for AI services with their personal credit cards today and generating apps out of spreadsheets they already use on a daily basis. While they may not have formed an intuition as to how many tokens equal what level of functionality, their quick mental math sees a $200/year Claude subscription to be worth anything between 1 and 7 FTEs. This is literally a 99% cost saving for 1 FTE alone.

Don’t shoot the messenger. One such business leader is the CEO of an accountancy and HR firm, and a close acquaintance.

Time

Waiting a sprint (two or three weeks) for business leaders or their subordinates (product owners, and so on) to witness the implementation of a bunch of user stories (with the promise that they’ll eventually amount to a finished product) doesn’t fly any more.

While Agile killed wireframes and prototyping with the pretense that only “working code” made sense, the business community never felt they really had a shippable product at any time during the sprint sausage-machine process. Now they want the entire agile sprint charade to be gone for good. “AI can do that in minutes rather than weeks” is their call.

I feel enormous anxiety confessing that business leaders expect AI-generated apps to get deployed straight into production. Business leaders reject the notion that AI-generated code is just for prototyping or validation. The value proposition is so good (i.e., extreme FTE reduction), that all cries about bugs, defects, and risks should be ruthlessly silenced somehow; “this is now unacceptable Luddite behaviour” is their response.

As an architect, I feel I need a complete, lengthy essay to elaborate on this point but I don’t see the business leaders backing out from this stance unless we all experience a production catastrophe that teaches everyone a lesson, which, as I elaborate further on, won’t dissuade them either.

Quality

This is perhaps the most contentious dimension in the value spectrum. There’s a huge discrepancy between how the business community and senior software engineers perceive AI-generated code.

What business leaders see:

• Something they can run and use straight away without asking IT
• Crucial common-sense features implemented out-of-the-box without asking for them explicitly (e.g., asking for a login screen doesn’t require the user to specify that the password input field comes after the email one, or that the widget is centered in the middle of the screen).
• Slick world-class UIs that abstract away from all the nerdy keywords such as SPA, CSS, React, state management, etc.
• All singing and dancing solutions with no time-wasting discrimination between “the backend”, “the database”, “the React component”, etc. It’s an app, not a mechanic’s car repair bill explanation!

Concerns that senior software engineers have:

• How to keep the application behaving in the same way when the model gets upgraded or when the inclusion of a new feature upsets the model’s interpretation of existing ones
• Contracts (APIs, tables, messages, etc.) with other applications, whereby these have been suggested by the model rather than being provided to it
• How to test the presence and absence of bugs without incurring an effort that negates the cost savings achieved by using AI-generated code in the first place
• Mapping prompt instructions to concrete code blocks for traceability, compliance, and verification purposes
• How does the model help (or doesn’t) in data migrations, whenever new generated code changes, deletes, merges, or splits columns or attributes found in the previous version.
• Finding a corner case that cannot be resolved using the prompt interface or code references, which results in human code augmentation based on a code base whose purpose is not to be extensive or maintainable; just the closest semantic approximation to the prompt.

As you can see, it’s not that business leaders and senior software engineers disagree; they are looking at completely different things.

Differentiation

The true differentiation game is only played by a handful of players who dominate the stock markets’ charts. Today, there isn’t much science about what a good website or app should look and behave like. What AI generators produce is in the league of your digital bank app or website: the smooth scrolling, the hamburger menu, the crispy icons; it’s all there.

As such, differentiation doesn’t lie any more in something that the AI generator may or may not do, but in the value of the underlying service:

• A savings account with a better interest rate
• A competitive energy rate during offpeak hours
• A two-hour delivery time (versus one or two days)

All worthwhile businesses have undergone ‘digital transformation’ already and implementing, say, Google Material Design, for example, doesn’t provide any edge whatsoever.

Vendor Risks and Sunk Costs

Onboarding an outsourcing vendor, and insourcing again, are traumatic business events. They take time and money, and contractual clauses may involve commitments of up to ten years. There’s also the element of transferring knowledge and capability to a third party, which is then quite painful to claw back.

This entire ordeal boils down to a $200 yearly bill, and a bunch of text stored using Windows Notepad. Again, I’m not playing devil’s advocate here; I’m just illustrating what the perception is at the moment.

Adoption Strategy

This is the major departure from how outsourcing to India took place. In the beginning, ‘shipping a job to India’ was literally done one employee at a time. Only much later it scaled to teams, departments, all the way up to entire business units.

This time, instead, early businesses adopting AI are going full monty instead. They are first deciding what the cost savings should be in advance and then shrinking their workforce accordingly (or placing hire freezes in the face of increased project demand).

What most people get wrong is that the layoffs are because of AI ’taking over their jobs’. AI hasn’t taken most jobs yet; instead, AI has captured the imagination of business leaders. In other words, it is not that employees have been made redundant following the automation of their jobs by AI. Instead, they’ve been made redundant in advance, based on the anticipation of AI being able to perform their jobs (and not just engineering!)

Some expect a major AI-production blunder which makes business leaders reconsider, but if the £10 billion IT disaster at the NHS did not put Accenture (and others) nor the NHS itself out of the business, we should hedge our bets. I don’t believe that an equivalent disaster in which AI is to blame would make most businesses bat an eyelid.

Conclusion

Indian outsourcers are today mature organisations and are employed based on capability rather than 1:1 FTE replacement. AI adoption (in the form of Vibe Coding) is, for now, seen through the lens of such early-days outsourcing mindset. Since we don’t yet know the exact shape that organisations need to adopt to make use of AI-generated applications in a sustainable manner, the debate, emotional polarisation, premature greed, and associated business mistakes will remain visible in the immediate future.

Before You Leave