AI has increased work, not replaced jobs at Policybazaar, says CTO Bibhu Krishna

TEXT START: “About 90% of the code that we generate today is written using AI,” Krishna said.

The Dissection

This is a local productivity report being presented as evidence against job displacement. Policybazaar has doubled or more than doubled its monthly production releases, cleared postponed work, expanded automation, and hired developers while using AI to generate roughly 90% of its code.

The structural reality is less comforting. AI has not preserved the old developer role; it has expanded the amount of output expected from each developer and shifted human work toward architecture, integration, verification, exception handling, and maintenance. “Two agents generate and check, then a lead reviews” is not proof that developers remain economically indispensable. It is a diagram of labor compression.

The company is using productivity gains to consume latent demand. That can increase hiring during an expansion phase because cheaper production creates more projects. It does not prove that human labor remains necessary once the backlog, product scope, and institutional constraints are exhausted.

The Core Fallacy

The central error is equating “no layoffs” with “AI has not replaced jobs.” Replacement can occur without immediate headcount reduction. A firm can retain the same workforce while each worker performs a smaller fraction of the total production process, while hiring less than it otherwise would have, or while converting future labor demand into software output.

The second error is confusing responsibility with execution. Krishna’s claim that “AI can just replace the skills, not the responsibilities” describes the current liability structure, not a permanent economic law. Human responsibility may remain formally assigned while the underlying cognitive labor is automated. A human signing off on an increasingly machine-produced system can become a legal and institutional lag defense rather than a durable source of mass employment.

The article shows P1 already operating locally: AI dominates code generation. It does not disprove P2 or P3. Human oversight persists because of current risk, accountability, and organizational design—not because human-only production has been proven economically superior.

Hidden Assumptions

• Productivity gains will continuously create enough new projects to absorb displaced cognitive labor.
• The backlog is effectively unlimited rather than a temporary reservoir of postponed work.
• Human review will remain mandatory even as verification systems improve.
• Accountability must continue to require substantial human cognitive labor rather than nominal approval.
• AI-generated systems will remain too context-dependent for deeper automation.
• Increased releases represent durable economic expansion rather than accelerated exhaustion of available work.
• Hiring today predicts employment demand after AI tooling matures.
• Customer empathy and institutional trust require humans to perform the underlying work, rather than merely to provide a human-facing interface.

These are not conclusions established by the article. They are the scaffolding holding its reassurance upright.

Social Function

Primary classification: partial truth functioning as transition management.

The CTO is likely describing the present accurately: AI has increased throughput, unlocked deferred initiatives, and created additional engineering demand inside this company. But the article turns a transitional labor-market response into a systemic conclusion. It reassures workers and institutions that automation is additive, while quietly documenting the mechanism by which human labor is being reduced to supervision, accountability, and housekeeping.

It also performs elite self-exoneration. Management can claim that nobody was replaced because nobody was fired, even as the production system becomes increasingly machine-dependent and future labor requirements are compressed.

The Verdict

This is not a refutation of the Discontinuity Thesis. It is a clean example of its lag phase.

Policybazaar is using AI’s productivity surplus to expand output before the economic system has finished absorbing the consequences. The company has not demonstrated that developers remain permanently necessary; it has demonstrated that AI can make a smaller amount of human expertise control a much larger production machine.

The statement “AI has increased work” is true in the short term and strategically misleading in the long term. Work volume can rise while human economic necessity falls. The backlog is being cleared by AI-assisted labor. Once the backlog, expansion opportunities, and human-liability requirements stop absorbing the surplus, the same system will require fewer people to produce more. That is not job preservation. It is the early administrative phase of obsolescence.