The family Macs were excellent machines with one critical defect: they were short on games. For a curious kid in the mid-1990s, the entertainment had to come from somewhere else. Rustam Lalkaka found it in the wires. He befriended the IT staff at his elementary school, accepted their broken computers, and assembled what he later called a “Frankenstein farm of busted Mac power PCs” at home. Some of it lived under his bed.
The machines spoke AppleTalk. Lalkaka listened. He explored SMTP servers, discovered the comic potential of spoofed email, and once sent his grandfather a message that appeared to come from Bill Gates. The prank is charming because the career it foreshadowed became so serious: products that route traffic, protect networks, and help teams understand what their software is doing while customers are using it.
“What I do at work today is not that different from building crazy AppleTalk networks under my bed,” he said during a 2020 Cloudflare TV conversation. The remark contains the useful version of a founder origin story. Curiosity survived contact with responsibility. The broken computers became internet infrastructure, then production agents, while the basic pleasure stayed intact: connect the pieces, see what happens, improve the system.
A career in invisible consequences
Lalkaka studied economics at the University of Wisconsin-Madison and added computer science. His early work crossed the usual boundaries between engineering, product, and entrepreneurship. Public biographies place him at Microsoft, Spire Health, and the startup Anapsis before Cloudflare. A 2014 Microsoft patent application named him among the inventors of a system for validating data transformations from end to end. Even then, the topic was verification: did the information survive the journey correctly?
At Cloudflare, where he spent seven years and eventually became vice president of product, the journeys grew global. In 2017 he introduced Argo, a smart-routing product built around a plain observation: the internet is a collection of networks joined by fiber, copper, microwaves, and trust, and the obvious route is not always the quick one. Argo measured conditions and selected better paths.
The following years produced a striking run of infrastructure launches and explanations. The Bandwidth Alliance coordinated cloud providers willing to reduce data-transfer fees for shared customers. Cloudflare's private backbone moved traffic across its own links. Magic Transit extended the company's network to protect and accelerate all IP traffic, not only web requests. Cloudflare One joined networking and security services into a common architecture. Lalkaka also wrote about HTTP/3, mobile acceleration, search performance, and Apple's iCloud Private Relay.
Public patent records name Lalkaka on granted inventions spanning data validation, VPN routing, compromised-origin detection, and early informational responses.
The common thread was hidden machinery with visible consequences. A route changes; a page arrives sooner. A relay separates a user from an identifying address. A network service absorbs an attack. Product management in that world is less about arranging buttons than making technical systems legible enough for customers to trust.
It also brings a specific kind of humility. Internet infrastructure fails publicly. Lalkaka later wrote that he drafted Cloudflare's first incident-management policy by borrowing Justice Potter Stewart's famous test for obscenity: “you know it when you see it.” It is a joke with operational teeth. Definitions help, but a live incident eventually refuses to remain a taxonomy exercise.
The dashboard gives up halfway
After enough incidents, Lalkaka and Achille Roussel became dissatisfied with the established observability bargain. Teams paid to collect logs, metrics, and traces. They built dashboards and alerts. Then, at the moment of consequence, a human still had to connect a code change with a customer complaint, determine whether an anomaly mattered, find the responsible service, and work out a fix.
The details annoyed them. Why was error rate difficult to slice by customer and deployment? Why could a log spike not be correlated cleanly with a configuration change? “Cardinality and cost” had become the shrug that ended the conversation. Their complaint was economic as well as technical: vendors earned more as customers ingested more data, while customers cared whether the software worked.
They founded Firetiger in 2024. Its agents combined three kinds of context that usually lived apart: the codebase, production telemetry, and the business outcome a team wanted. A human could define the destination. The agent would decide what to measure, watch the system, investigate deviations, recommend a change, and check the result.
The production feedback loop
Then coding agents improved. Firetiger had been designed to reduce operational toil, but the sharper opportunity came from a change in software supply. Claude Code, Cursor, and their peers made it much easier to produce code. Lalkaka watched his own team ship more, with larger changes and less line-by-line human familiarity. Risk accumulated downstream. Review, deployment, and production verification became the constraining steps.
His formulation is crisp: AI made writing code cheap, while shipping code remained expensive. The practical measure of progress is not tokens consumed or pull requests opened. It is whether useful changes reach production faster, recover cleanly when they fail, and improve outcomes without increasing the failure rate.
Make the loop prove itself
Firetiger's most persuasive demonstrations came from turning the product inward. The company runs hundreds of language-model agents, making prompt caching a meaningful line item. A cache can save money when prompts repeat, but the wrong time-to-live can make writes cost more than later reads recover. Workloads keep changing, so a perfect static setting expires almost as soon as it is calculated.
The team gave a Firetiger agent one objective: reduce prompt-cache waste. Each day it examined usage and billing data, calculated read-to-write ratios, recommended code or configuration changes, and returned after humans approved the fixes. Some caches needed longer lives. Others needed shorter ones. Some needed to disappear. Firetiger reported a 77 percent reduction in inefficient cache spending.
The number is useful, though the method is the larger point. A human could run the same campaign by hand. That person would inspect every agent and deployment, calculate the economics, draft changes, and repeat the exercise as traffic shifted. Firetiger moved the recurring attention into software and kept human approval at the decision point. It automated the campaign, not merely the calculation.
Change Monitors applied the same structure to pull requests. An agent reads a diff during review, studies the affected systems, and prepares a monitoring plan. After deployment, it checks the relevant production signals and comments on the pull request with a verdict. The code review gains an epilogue: here is what the change actually did.
Six public months, years of accumulated context
Firetiger publicly launched on February 18, 2026, announcing a $7.6 million seed round led by Sequoia Capital. The investor list included Cloudflare co-founder Matthew Prince, Segment co-founder Calvin French-Owen, former Formula One champion Nico Rosberg, and former Amazon executive Jeff Wilke. The company sold agents rather than data ingestion, with the commercial model reflecting the product argument.
On August 13, the story compressed again. Firetiger announced that its team was joining Cursor. The pairing was structurally neat. Cursor's agents create code; Firetiger's agents observe what happens when code meets production. Cursor said Firetiger's work would appear across its product as the company developed a path from writing software to running it, including Change Monitors that flag problems after deployment.
For Lalkaka, the destination sounds like a scaled-up version of the network under his childhood bed. The machines are more capable, the stakes are higher, and the messages are no longer fake notes from Bill Gates. Yet the pleasure of the work still comes from connection. Code talks to telemetry. Telemetry talks to an agent. The agent talks to another agent or asks a human to approve a fix. The result flows back into the system as evidence.
His aspiration is a software factory where agents can write, verify, investigate, and improve, while people steer toward outcomes that matter. The phrase “human-less” appears in his description of that future, but his own product examples keep humans at consequential joins. They click Fix. They merge. They decide what matters. Autonomy handles the repetition; judgment supplies the direction.
There is a builder's lesson hiding in that division of labor. Tools become valuable when they carry work across the point where the old workflow stopped. The alert should lead to an investigation. The investigation should become a change. The change should face a measurement. Every handoff is a chance for the loop to break, and therefore a place to build.
Lalkaka has followed that logic from networks to product teams to AI agents. Make the invisible visible. Shorten the trip between action and consequence. Keep enough wit around to survive the incident call. The broken Macs have left the bedroom. The feedback loop is still running.