For decades, India and the Philippines built massive industries around what is called Business Process Outsourcing, or BPO. Workers in Manila and Bangalore read messy emails, police reports, doctor's notes, and invoices, and type the relevant information into corporate databases. They function, essentially, as a human bridge between unstructured data — the random documents the world produces — and structured data, the clean spreadsheet rows that businesses can actually act on. The Philippines' BPO sector alone employs one-point-eight-two million people and generates eight to nine percent of national GDP.
Modern AI models can now perform that conversion in under a second for approximately four cents per task. A human worker in India performing the same task costs roughly two dollars. The cost gap is not a small efficiency improvement. It is a fifty-fold collapse in unit economics. Across just two use cases — banking trade breaks and insurance claim parsing — AI could displace an estimated two-point-eight-five billion dollars in annual human labour at a compute cost of only twenty-six million dollars. That is roughly one hundred and forty-two thousand jobs at the boundary between machine and human work.
The disruption resembles a gold-rush principle applied to AI. When everyone is mining for gold, the people who get rich are not the miners but the ones selling shovels. AI model providers like Anthropic, OpenAI, and Google capture less than one percent of the labour value they destroy. Most of the savings flow to the companies that buy the AI services and to the consumers who eventually pay lower prices. But because the model providers sit underneath every industry, the small slice they keep adds up to billions across the economy. A useful framework called the 93/7/2 split divides the work this way: ninety-three percent of back-office tasks are already automated, seven percent currently require human judgment, and AI will likely shrink that human share to roughly two percent.
The pain hits asynchronously because of how IT contracts are structured. Most major outsourcing deals run for three to five years. The pricing pressure shows up first, with clients demanding twenty to forty percent rate cuts at renewal. Layoffs come later, when contracts roll off in 2027 and 2028 and the new staffing requirements look nothing like the old ones. Most observers assume that voice AI is the bigger threat to outsourcing — calls being handled by AI agents instead of human ones. The author argues, contrarian to that view, that non-voice work, the supposedly safe fallback, is actually more vulnerable.
For students considering careers in technology, finance, consulting, or any office-based field, this is a forced reality check. The work that once required armies of entry-level analysts — reading documents, extracting data, filling forms — is the first thing AI eats. The jobs that survive will require judgment, client relationships, or the ability to build the AI workflows themselves. For young people in India, the Philippines, or anywhere with a large outsourcing economy, the macroeconomic stakes are personal: currency stability, urban real-estate prices, and the family remittances that millions of households depend on. The countries that pivot from selling labour to selling AI-augmented judgment will adapt. The rest face a much harder transition than their political systems are currently prepared for.
An AI model costing four cents per task is about to vaporize $2.85 billion in human labor — and the company selling that AI captures less than 1% of the value it destroys.
For decades, India and the Philippines built massive industries around what's called Business Process Outsourcing, or BPO. Workers in Manila and Bangalore read messy emails, police reports, doctor's notes, and invoices, then type the relevant information into corporate databases. They are, essentially, a human bridge between unstructured data (random documents) and structured data (clean spreadsheet rows).
AI models like Anthropic's Claude Haiku can now do this conversion in under a second for about four cents per task. A human in India costs roughly $2.00 per task. The author argues that across just two use cases — banking trade breaks and insurance claim parsing — AI could displace $2.85 billion in annual human labor while costing only $26 million in compute. That's roughly 142,500 jobs.
Think of this as the gold-rush principle applied to AI: when everyone's mining for gold, the people who get rich aren't the miners — they're the ones selling shovels.
If you're considering a career in tech, finance, consulting, or any office-based field, this is your reality check: the work that once required armies of entry-level analysts — reading documents, extracting data, filling forms — is the first thing AI eats. The jobs that survive will require judgment, client relationships, or the ability to build the AI workflows themselves. For students in India, the Philippines, or anywhere with a large outsourcing economy, the macroeconomic stakes — currency, real estate, family remittances — are personal.
Historically, every automation wave — looms, ATMs, spreadsheets — destroyed specific jobs but created new categories of work. The open question is whether AI follows that pattern or compresses the timeline so brutally that whole national economies can't adjust. Watch for second-order effects: real estate prices in Makati and Bangalore tech corridors, the rise of 'AI wrapper' startups charging $1-2 per task for what costs them $0.04, and political pressure on governments to regulate AI deployment in services. The companies — and countries — that pivot from selling labor to selling AI-augmented judgment will survive. The rest become Gaitonde.