How Cars24 Measures the Code That Reaches Production
As AI-written code accelerated across Cars24's 400-person engineering org, commit and PR counts stopped telling them what actually shipped. With Macroscope Status, Cars24 measures output by what lands in production—and tripled one team's production-landed code in three months.
Cars24 is one of India's largest auto-tech companies, with more than 300 million users globally, more than a million cars transacted, and more than $1 billion in revenue. It operates in 250 cities across India, the UAE, and Australia. On peak days, Cars24 completes one transaction every second.
Solving for scale is a core operating principle of Cars24, and it governs how the company builds software as much as how it sells cars. Over the last several years, Cars24 has become one of India's most advanced adopters of AI. Its monthly financial close has gone from 15 days to 3, service-request turnaround has fallen from 32 hours to 30 minutes, and the large majority of its code is now written with AI—and all of it is reviewed by AI.
What began as an internal AI transformation recently became a product: in August 2026, Cars24 launched Deployment Inc, an independent AI company that helps other businesses deploy AI in their operations, using Cars24's own AI rollout as its playbook.
Inside Cars24's own engineering organization, its rapid AI adoption raised a question: was the team actually shipping more to production, or just generating more code?
The Limits of Traditional Productivity Metrics
As coding agent adoption accelerated across Cars24's 400-person engineering organization, the company needed a clearer way to separate coding activity from actual delivered outcomes.
Commit volume and PR activity, the measures they had relied on historically, could show that engineering work was happening. Neither indicated whether that code ultimately reached production. At the same time, token consumption was growing rapidly across the business, creating a challenging question for its engineering leaders: how much output was that token spend actually producing?
"How do you measure productivity when token usage no longer tells you what work is actually getting done?" — Jayesh Gupta, AI & Innovation, Cars24
A Signal for Production-Landed Code
Cars24's solution was to change what it measured. The company set out to rebuild productivity measurement around outcomes rather than activity, building a centralized dashboard that pulls signals from across the business. Since December 2025, Macroscope has been the source of truth for engineering output, measuring the code the team produces and how much of it reaches production.
"We pull every signal we can into a central dashboard. Macroscope is one of those sources. Right now, it's the strongest contributor." — Jayesh Gupta, AI & Innovation, Cars24
Macroscope Status gives Cars24 two complementary perspectives:
- Estimated coding time. Macroscope estimates how long a code change would have taken to write manually. Cars24 uses these estimates directionally to identify where work is concentrating.
- Whether code lands. Macroscope distinguishes "pushed" from "landed" coding time, so Cars24 can measure output by what shipped to production and see what proportion of their code actually shipped.
With insights from Macroscope Status, Cars24 found teams with high commit volume but very little of that work reaching production. Once they could establish where work was consistently getting stalled, they were able to address the underlying causes. Cars24 piloted Macroscope with one of its fastest-moving teams, where code volume had grown faster than the path to production. Within three months, Macroscope helped triple the share of that team's code reaching production.
"The most actionable signal for us is seeing what code lands in staging versus production." — Jayesh Gupta, AI & Innovation, Cars24
Why "Landed" Matters Beyond Engineering
Cars24's finance team capitalizes software development costs, amortizing the engineering time and token spend that go into long-lived platforms. This is ordinary accounting practice, but it depends entirely on establishing which engineering work actually reached production.
"You can only capitalize a building once it's built, not while it's still being drawn. Software works the same way, so we go by what landed." — Jayesh Gupta, AI & Innovation, Cars24
Code Review at Agent Scale
In a large company where agents write the majority of code, PR volume has far outpaced what Cars24's engineers can review.
Cars24's engineering organization relies on Macroscope Code Review as the first line of defense: catching bugs, flagging risky changes, and suggesting fixes before an engineer gets to a PR, so human reviewers arrive with confidence that what reaches them is ready for their attention.
Cars24 has also adopted Macroscope Approvability so that its engineers can get a clear signal about whether a code change is safe to merge or warrants extra human judgement.
"We feel like Macroscope has solved code review." — Jayesh Gupta, AI & Innovation, Cars24
At this scale, code review is not just about finding bugs: it's about knowing which changes are safe to trust, which need attention, and where human reviewers can best spend their time.
The Lesson for Large Teams Adopting Coding Agents
For leaders of large engineering teams deploying coding agents at scale, Cars24's experience points to a broader insight: the activity metrics that worked before agents no longer tell you what's getting shipped.
With Macroscope, Cars24 can analyze its coding output with production-landed data—so leaders can understand not only how much engineering work is happening, but how much of it actually reaches production.
