Infrastructure That Thinks
Razorpay’s real problems, agentic bets, hiring trouble, and what Swiggy reveals about the cost of speed
Abhinav Jain | Breath Hacker
A Breath Hacker investigation into two companies running as fast as their infrastructure will allow, and what breaks when the ground shifts.
Two companies. One processes money for twelve million businesses. The other delivers dinner to half of urban India. Both are building at speed. Both are burning. And both are making strategic bets that look like certainty from the press release and like philosophy from the inside.
The Breath Hacker lens does not ask what a company announced. It asks what a company believes about itself, one layer below the announcement. This is that read.
PART ONE: RAZORPAY
01 The Problems Nobody Puts in Press Releases
Problem 1: The Transparency Gap
Razorpay processes payments at a scale that would have seemed fictional a decade ago. But scale did not fix the problem its merchant base has been describing for years: you do not know what you are paying, why your settlement is delayed, or what the root cause of any given failure is.
Support responds with generic messages. Fee breakdowns require hovering over obscure UI elements. Banks and partner system failures produce no transparent report, just silence.
This is not a small product problem. It is a trust infrastructure problem. Every business using Razorpay has money sitting inside a system they cannot see clearly. When that system hiccups, the merchant absorbs the uncertainty. That asymmetry erodes the relationship faster than any competitor offer.
Opacity is not a support failure. It is an architectural choice that gets made by default when you are building fast. Razorpay built fast. The opacity was the price.
Problem 2: The PMLA Overhang and Regulatory Risk
Razorpay spent years under an Enforcement Directorate money laundering probe, with assets worth Rs 689 crore frozen or attached at peak. The Supreme Court ultimately upheld the Karnataka High Court order quashing the case in March 2026, but the damage to reputation and the management bandwidth consumed during that period cannot be quantified in a filing.
The underlying issue, that payment aggregators sitting between merchants and the financial system are structurally exposed to the illegal use of those merchants, did not go away with the court order.
Every payment aggregator at Razorpay’s scale faces this structural exposure. The question is not whether a bad merchant will slip through, but how fast you catch it and how cleanly you have documented your due diligence. That documentation problem is an engineering problem as much as it is a legal one.
Problem 3: The RBI Ban Period and Its Real Cost
Between 2022 and 2023, the Reserve Bank of India banned Razorpay from onboarding new merchants. The official narrative afterwards positioned this as a period of diversification and product deepening. That is partially true. But what also happened was that competitors onboarded the merchants Razorpay could not. Market share that takes years to build can shift in months when one player is operationally blocked.
Problem 4: The FY25 Loss and What It Actually Means
Razorpay reported a Rs 1,209 crore net loss in FY25. The company attributed this to one-time costs from its reverse flip, the process of re-domiciling the parent entity from the US to India ahead of its IPO. That framing is accurate. But it is not complete.
The reverse flip itself was a strategic necessity, not an operational accident, and it cost the company a tax overhang that will affect the IPO valuation conversation. The target IPO raise of Rs 5,000 to Rs 6,000 crore at a notable discount to the 2021 Series F valuation of 7.5 billion dollars is the financial footprint of that strategic decision.
Problem 5: Merchant Support at Scale
Twelve million merchants is not a support team’s problem. It is an automation problem that somebody has not fully solved yet. Reviews across Shopify and merchant forums reveal a consistent pattern: widget failures, settlement confusion, Shopify integration breakdowns, and support interactions that cycle through email without resolution.
This is the merchant experience at the edges of the product, where the sales team has left and the account management has not yet arrived.
02 How Razorpay Is Answering with Agentic AI and Python
What was announced at FTX 2026, and what it actually means for the infrastructure.
At its flagship FTX 2026 event in March, Razorpay launched what it called the world’s first AI-native Agent Studio for payments, built on Anthropic’s Claude Agent SDK. Alongside it came the Agentic Experience Platform, a conversational interface layer designed to replace traditional dashboards for merchant interactions.
Eight specific agents shipped at launch: a Dispute Responder that handles chargeback cases autonomously, a Subscription Recovery agent (with ElevenLabs voice) that contacts customers to recover failed payments, two Abandoned Cart Conversion agents operating inside Zomato and other platforms, a Cashflow Forecaster projecting positions three to seven days ahead, an RTO Shield and RTO Insights for return-to-origin prevention in ecommerce, and a Settlement Insights agent. A no-code Agent Builder was released in beta for finance and operations teams.
Razorpay did not just add AI to its product. It rebuilt the category assumption: that payment infrastructure is passive. The Agent Studio bets that infrastructure should think, not just process.
The technical architecture is worth understanding. Each agent runs inside a Claude reasoning loop: observe signals from Razorpay’s payment data, reason over what action to take, execute via Razorpay’s 400 plus APIs and third-party integrations, report back.
The consent guardrails are explicit: the agent never sees raw financial data beyond what is necessary for its specific task. Razorpay’s CPO Khilan Haria stated the company evaluated multiple AI providers and chose Claude for its reasoning capabilities and suitability for high-stakes financial workflows where errors have direct monetary consequences.
The Python dimension matters here. Razorpay’s payment node integrations now include one-click deployment for n8n, Replit, and Vercel. Python-native workflow builders can connect Razorpay payment logic directly into agent pipelines.
The no-code builder reduces the engineering barrier for operations teams who understand business logic but do not write code. This is an important architectural shift: Razorpay is moving from a tool developers integrate to a platform that business teams can configure.
The agentic payment concept that Razorpay first introduced at the Global Fintech Festival 2025 with NPCI and OpenAI, showing AI completing payments rather than just recommending them, has now expanded to in-app commerce with Zomato, Swiggy, PVR Inox, and others.
The vision is a world where a customer’s conversation with an AI produces a completed transaction without navigating a checkout flow. Razorpay is betting that it owns the payment rail for that future.
The Agentic Experience Platform replaces the traditional merchant dashboard with a conversational interface. Finance teams can upload reconciliation documents and complete what previously took hours in seconds. Merchant onboarding is simplified to a conversation.
The company calls this “agentic onboarding.” The honest translation: the KYC and compliance complexity that caused so much merchant friction is now handled by an AI that can hold context across a long document, ask the right follow-up question, and not lose patience.
03 The Hiring Problem Nobody Is Naming Clearly
Razorpay’s hiring data tells a story the press releases do not. Active job postings fell 26.7 percent in 2025, down to 107, from 146 in 2024. Hiring velocity slowed from 148 new roles per month to 95. The company grew headcount from around 3,988 to 4,208, a 5.4 percent increase, but the trajectory is decelerating.
The company has 38.6 percent of its workforce in engineering. It is building an Agent Studio on a cutting-edge AI SDK. It is targeting an IPO while managing a significant product rebuild. And it is hiring for senior engineering roles that its own recruitment partners have described as staying open for longer than expected because of unmatched complexity and scalability requirements.
The problem is not whether Razorpay can find engineers. The problem is whether the engineers it finds can hold two things at once: the precision required for high-stakes financial infrastructure and the fluency required for AI-native product thinking.
That combination is genuinely rare. Most engineers who are strong on distributed systems and payment reliability are not native to agent architectures. Most engineers who are fluent in LLM pipelines and Claude SDK integration have not spent years thinking about settlement reconciliation, PMLA compliance, and fraud detection at scale.
Get Abhinav Jain’s stories in your inbox
Join Medium for free to get updates from this writer.
Razorpay is asking for both. That gap is real, and the slowdown in hiring velocity suggests the company has not yet found a clean sourcing path for it.
The RazorpayX Payroll Report for 2026 noted that across its customer base of 6,500 plus startups, AI-powered roles grew by 96 percent. The company is watching that hiring trend from its own payroll product while trying to win the same talent war. That is both an insight and a liability. Razorpay knows the market for AI talent better than almost anyone. Knowing it does not mean you can outcompete for it.
The narrative logic here, read through the Breath Hacker lens, is this: Razorpay built its brand on being the infrastructure that developers love. That brand recruited engineers who wanted to build for builders. The shift toward agentic AI and conversational commerce changes who the customer is.
When your customer is a finance manager configuring an agent through a no-code builder, the engineering identity of the company changes. That transition is disorienting for teams that came in because they wanted to build payment APIs, not agent orchestration platforms.
The hiring problem is not a sourcing problem. It is an identity problem that manifests as a sourcing problem.
PART TWO: SWIGGY
04 What Is Actually Broken at Swiggy in 2026
Revenue at record highs. Losses at record highs. Both things are true simultaneously.
The Profitability Paradox
Swiggy reported quarterly revenue of Rs 5,561 crore in Q2 FY26, its highest ever. The net loss for the same quarter was Rs 1,092 crore. For full-year FY26, cumulative losses since inception crossed Rs 15,000 crore. The stock, which listed in late 2024, traded at roughly Rs 327 in January 2026, down 30 percent from its 52-week high. It slipped below its QIP issue price of Rs 375 in early 2026 amid heavy selling.
The company is growing revenue at over 50 percent year on year while simultaneously losing money on every quarter’s operations. This is not a rounding error. It is the foundational business model tension that everything else flows from.
Instamart: The Expensive Bet That Must Win
Swiggy’s quick commerce arm, Instamart, is the company’s most critical and most costly product. It runs on dark stores, which are warehouses optimised for ten-minute delivery rather than customer foot traffic. Each dark store has capital costs, staffing costs, and inventory costs.
At 40 percent utilisation, Instamart is losing money. Even as utilisation improves, the fixed cost base of the dark store model means the path to breakeven is long.
Blinkit, Zomato’s quick commerce product, now commands roughly 45 percent of the Indian quick commerce market. Zepto is expanding aggressively with premium positioning.
Swiggy is in third place in a sector it helped create. JM Financial warned that Swiggy needs a fundraise of over 500 million dollars for its quick commerce ambitions. The Rapido stake sale, which Swiggy executed to raise cash, offers only about two quarters of runway by some analyst estimates.
Swiggy is not losing to a better product. It is losing a capital race in a market where capital determines shelf space and shelf space determines customer habit. The product is fine. The bank account is the problem.
Delivery Cost Pressure
A Rs 3 per litre petrol price hike in mid-2026 pushed delivery costs higher across the industry. For Swiggy, which operates one of India’s largest delivery fleets, this is not a footnote. Delivery costs are the variable cost that determines unit economics.
When fuel goes up, the contribution margin on every order narrows. When the contribution margin narrows and the marketing spend required to hold market share stays high, the path to operational breakeven moves further out.
Swiggy also faced a delivery partner shortage in a quarter affected by monsoon disruption. That shortage reduced delivery capacity at the exact moment when demand was high. The gap between demand and fulfilment erodes customer trust faster than a poor app experience.
The Strategic Retreat Question
Swiggy’s leadership signalled in mid-2026 that it would not engage in an all-out spending war against Blinkit and Zepto. That restraint is financially sensible. It is also strategically risky. In a market where dark store density determines delivery time and delivery time determines customer retention, the company that does not expand its network at pace loses the ground it cannot easily recover.
Swiggy faces a middle-ground trap: not aggressive enough to catch Blinkit, not differentiated enough to command a premium.
The quick commerce sector in India is also being crowded from a different direction. BigBasket, backed by Tata, is expanding. Amazon has entered rapid retail. Swiggy Instamart is now fighting on multiple fronts: the pure-play speed competition with Blinkit and Zepto, and the breadth-and-trust competition with established grocery brands.
New Bets, Unclear Timelines
Swiggy is incubating several new products: Pyng (a professional services marketplace), SNACC (a social commerce product), Swiggy Sports, and others under a Platform Innovations umbrella. These are genuine diversification bets. They are also early, loss-making, and competing for management attention at a moment when the core business requires it.
Each new vertical is a bet that Swiggy’s logistics and consumer data infrastructure transfers into adjacent categories. The thesis is sound. The timing is aggressive.
05 The Breath Hacker Read: What Both Companies Reveal
Here is the layer below the layer.
Razorpay and Swiggy are both Indian technology companies at a moment of institutional maturation. Razorpay is filing for an IPO. Swiggy completed one and is now accountable to public market expectations. Both are making large, expensive bets on infrastructure that does not yet produce returns at the scale required. And both are doing this while managing the identity transition that comes when a startup becomes an institution.
The Razorpay Agent Studio is a bet that agentic AI will become the primary interface through which businesses manage financial operations. If that bet is correct, Razorpay owns the rails and the agents. If it is premature, they have built a sophisticated product that their merchant base is not ready to use.
Swiggy’s Instamart is a bet that Indian consumers will pay a convenience premium sufficient to support a dense dark store network. That bet requires capital Swiggy is struggling to accumulate at the rate the bet demands.
Both bets are structurally sound. Both are early by at least one cycle. The question is not whether the vision is right but whether the runway is long enough for the vision to become habit before the money runs out.
The Breath Hacker reading of both companies is this: they are doing what ambitious infrastructure companies must do, which is to spend ahead of the behaviour they need to create. Razorpay needs merchants to trust agents with their payment operations. Swiggy needs consumers to trust ten-minute delivery with their weekly grocery runs. Both of those trust transfers take longer than a product launch cycle.
The hiring problem at Razorpay and the funding problem at Swiggy are the same problem in different forms: you cannot build the future without the people who understand it, and you cannot attract the people who understand it without showing them a future that is already arriving. That circularity is the real infrastructure challenge. Not the technology. The belief.
Razorpay is building the right infrastructure for the right future and has the regulatory clearance, the revenue growth, and the Anthropic partnership to make a credible claim. Its problems, transparency gaps for merchants, a complicated IPO valuation story, senior engineering talent scarcity, and a product identity transition from developer tool to agentic platform, are real and are not solved by a press release.
The Agent Studio is a genuine architectural bet, not a feature launch. Whether it earns trust faster than the transparency and support failures erode it is the real product question.
Swiggy is a company winning on revenue and losing on margin, third in the segment it pioneered, and making diversification bets before its core bet is resolved. The middle-ground position in quick commerce is the most dangerous: not fast enough to win on speed, not profitable enough to win on patience.
Swiggy’s 2026 story is about a company that correctly identified where Indian consumer behaviour was heading and arrived there slightly behind the capital it needed to hold the position. Both companies are doing serious work. Both are running on timelines that require things to go right in a specific order.
Infrastructure that thinks faster than the humans who trust it does not automatically earn that trust. The gap between what a system can do and what a merchant, consumer, or investor believes it will do reliably is not closed by a demo. It is closed by years of unglamorous consistency.
Razorpay’s agentic leap and Swiggy’s quick commerce war are both correct visions executed against a culture that adopts slowly. The founders who survive this phase are not the ones with the best technology. They are the ones who understood that trust is the product, and everything else is the delivery mechanism.
Breath Hacker | sirabhinavjain13.substack.com
