TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Software engineer Neil Kakkar says he merged 11 times as many pull requests per month at AI-native startup Tano as he had at PostHog, based on his own work history. His account describes both faster output and new pressures, including context switching, reduced shared understanding and concern about AI usage limits; it is one engineer’s experience, not an industry-wide study.
Software engineer Neil Kakkar says he merged about 11 times more pull requests per month at AI-native startup Tano than at PostHog, where he worked before the current generation of coding agents became part of his workflow. His first-person account, published by The Pragmatic Engineer, describes a shift in how software work gets done—and the new coordination costs that can accompany faster output.
Kakkar reports that at PostHog he typically merged between eight and 20 pull requests a month, while at Tano he has been merging more than 200. He describes the comparison as roughly 11 times the monthly pull-request volume. He cautions that pull requests are an imperfect measure and says he was not trying to maximize the count. The figures reflect his own work, not a controlled study or a measure of software quality.
The comparison spans different roles and periods. Kakkar spent four years at PostHog, where he worked across analytics, experimentation, feature flags and other products. At Tano, he joined as a founding engineer and says he completed comparable breadth of work in eight months, including an affiliate platform, improvements to a content-monitoring system and a migration of the company’s agent architecture. He also cites a rough comparison with Bloomberg, where he began his career, but says those figures are less detailed.
He attributes much of the change to AI tools and the way Tano organizes engineering work. At PostHog, he says, AI was largely used for code completion and repetitive tasks. At Tano, he joined after coding agents had become more capable. The faster pace has reduced some delays associated with communication and review, in his account, but brought other problems: more context switching, weaker shared context, less attention to design and concern about limits on AI plans.
Faster Shipping, New Coordination Costs
Kakkar’s experience offers a concrete, though individual, view of the shift from AI-assisted coding to agent-supported software development. If engineers can produce and merge changes more quickly, teams may be able to test ideas and respond to product needs at a faster pace. But a higher pull-request count alone does not show whether software is more reliable, whether customers benefit more, or whether the work required less total effort.
The account also points to a management challenge: speed does not remove the need for coordination. Kakkar says teams need people to adopt AI tools and experiment with how they work, while keeping enough shared understanding to make decisions and maintain systems. Without that, faster individual output can create more handoffs and make design choices harder to track.
For software workers and employers, the practical question is not simply how many changes an agent can produce. It is how teams review, test and understand those changes, and whether the gains in delivery are worth the additional costs in attention, tool usage and oversight.
Products Worth Considering
Bose QuietComfort Wireless Noise Cancelling Headphones, Black
Sony WH-1000XM4C (M4 2nd Gen) Wireless Noise-Cancelling Headphones, Black
Bose QuietComfort Ultra Earbuds (2nd Gen), Wireless Noise-Cancelling Earbuds, Black
As an affiliate, we earn on qualifying purchases.
From Tab Completion to Coding Agents
The essay was the winning entry in a challenge marking The Pragmatic Engineer’s fifth anniversary. The publication says it received 125 submissions and awarded Kakkar’s entry a $10,000 grand prize. It described him as a London-based engineer with nine years of professional experience, including work at Bloomberg, PostHog and Tano.
Kakkar frames his comparison as two different working eras. He treats his PostHog experience through late 2024 as a pre-AI baseline, while noting that PostHog has since adopted AI more extensively and that its own output is increasing. He describes Bloomberg as a much larger company with established processes, PostHog as a startup that grew substantially during his tenure, and Tano as a small, AI-native company where he was the first engineer.
Those differences matter when interpreting the numbers. The comparison is not only about software tools: the companies varied in size, culture, engineering practices and the length of time covered. Kakkar’s history provides a detailed personal case, but it cannot isolate the effect of AI from those other factors.
“I’m shipping 11x more PRs at Tano than at PostHog.”
— Neil Kakkar, software engineer and essay author
What the Pull-Request Count Cannot Show
The figures come from one engineer’s account, and the source does not provide an independent audit, a full methodology for counting each month, or a matched comparison controlling for role, project scope and company conditions. Kakkar says his Bloomberg numbers are rough, and his reported work at PostHog and Tano covers different time spans and organizational settings.
It is also unclear how much of the increase comes specifically from AI agents rather than from Tano’s smaller size, its engineering culture, Kakkar’s seniority or the products and migrations he worked on. The essay does not establish whether the higher volume led to fewer defects, better customer outcomes or lower total development costs. Kakkar also notes that his former employer PostHog has since become more active in AI, so his pre-AI comparison does not describe its current output.
Further Essays and Broader Evidence
The Pragmatic Engineer said it plans to publish a follow-up featuring runner-up essays and links to other submissions it found notable. Those accounts may offer additional experiences, but they would still not, by themselves, establish an industry-wide productivity effect.
For now, the evidence here is Kakkar’s retrospective comparison. Broader conclusions would require comparable data across more engineers and teams, alongside measures of quality, reliability, review effort and customer impact. Until then, his figures are best read as a specific case study of one engineer’s changing workflow, with measurable output claims and unresolved questions about what the count means.
Key Questions
Who reported the increase in pull requests?
Neil Kakkar, a software engineer who wrote the winning essay in a challenge published by The Pragmatic Engineer, reported the increase based on his work at PostHog and Tano.
How large was the reported change?
Kakkar says he went from roughly eight to 20 merged pull requests per month at PostHog to more than 200 at Tano, which he characterizes as about an 11-fold difference. The figures are his own account.
Does the report prove AI makes engineers 11 times more productive?
No. It compares one engineer’s output across two companies and different periods. It does not isolate AI’s effect or measure quality, reliability, customer outcomes or total effort.
What drawbacks did Kakkar describe?
He cited context switching, loss of shared context, less focus on design and concerns about AI plan limits as pressures accompanying the faster pace.
What is expected next?
The Pragmatic Engineer said it would feature runner-up essays in a later issue. The source does not specify when that follow-up will appear.
Source: newsletter.pragmaticengineer.com
