Case study · Leena AI · Enterprise virtual assistants · Series B, YC S18

Leena AI, ten months in.

Where usage stands ten months after rollout, and what it has been worth.

Get a demo

People on ProdE

166 93% have used it (155)

Questions in 10 months

15,272 in 3,615 conversations

Running coding agents

103 104,666 agent tool calls

Codebase

53 repos 7M+ lines of code

"ProdE is the only tool we've used that actually understands our codebase end to end."
Mayank Kapoor · SVP, Leena AI

Company profile

Leena AI builds enterprise virtual assistants for HR, IT and finance teams, serving clients in healthcare, FMCG, technology, aviation, telecom, retail, pharma and education. YC S18, Series B. Its product, engineering, QA, support, solution consulting and customer success teams, 166 of them on ProdE, work across 53 repositories and more than 7 million lines of code.

Then and now

Three job functions at three months, seven at ten.

	3 months in	10 months in
On the platform	93	166
Have used it	68	155
Teams	engineering, product, support	engineering, product, QA, design, solution consulting, customer success, support

What changed

Before	After (client reported)
Answer to a cross-team question in 2 to 3 days	15 to 30 minutes
Engineer onboarding in weeks	Days
Cross-service debugging in hours	Minutes
PRDs built on assumptions	PRDs built on the actual code

The before

Leena's solution consultants, customer success and support teams know the product and the clients but do not work in the code. A question about how a feature behaves for one client meant finding the right engineer, interrupting them, waiting, and often a second round when the answer did not quite fit.

Engineering had a similar problem across 53 repositories: the service named in a ticket was often not the one that owned the behavior.

Every one of those questions cost an engineer's attention and the asker a wait. Leena's teams asked 15,272 of them in ten months.

Who uses it, and for what

Team	How they use it
Engineering	Cross-service debugging. Finding which of 53 repositories owns a behavior. Coding agents that read the system map before they touch code. Fixes start in the right service. Senior engineers field fewer questions.
Product	Checking a PRD against the existing code: what the feature touches, what already exists, where the edge cases are. Engineering gets specs grounded in the code. Fewer surprises in development and QA.
QA	How a flow actually works and what validation applies, before writing tests or filing a bug. Tests match real behavior. Root cause is known before the ticket is raised.
Solution consultants	How a feature behaves for a specific client, configuration and integration questions during a deal. Answered on the call instead of the next day.
Customer success and support	Client tickets, data questions, the cause of an issue a client reported. Resolved without waiting on an engineer. Answers come with references into the code.
Design	How a flow is implemented today before redesigning it. Designs start from the real system.
New joiners	Learning the system by asking it questions. Productive in days rather than weeks.

Agents on the same map

103 people connect their coding agents to ProdE over MCP. 104,666 agent tool calls in ten months. The agent pulls the system map and the relevant code before it starts, so it works in the service that owns the behavior rather than the one named in the ticket. Agent use has grown steadily every month since launch.

What they ask

Objective	Share of questions
Understand the system (architecture, data model, where is X implemented or configured, which endpoint)	51%
Debug and root cause	20%
Plan and design a feature	12%

Daily use

Daily active users, monthly average. ProdE analytics dashboard.

Daily use grew about five times in ten months, from the November 2025 rollout to October 2026, across chat and coding agents.

What it adds up to

What this is worth modelled

~850 engineer hours a month returned: questions from other teams that no longer land on engineers, cross-repo digging replaced, agent sessions. Five engineers' worth of time. ~230 engineer interruptions a month that no longer happen. ~$850K a year at $80 an engineer hour.

Not counted: the waiting time on every question, and onboarding time.

Counts are from ProdE's analytics. Hours are modelled per question and per agent session from how a question got answered before ProdE: find the owner, interrupt them, they dig and answer, a clarification round.

The results

Metric	Result (as of Oct 2026)
People on ProdE	166, 155 have used it (93%)
Codebase	53 repos, 7M+ lines of code
Questions answered	15,272 in 3,615 conversations
Who uses it	Engineering, product, QA, design, solution consulting, customer success, support
Agents on ProdE	103 people, 104,666 agent tool calls
Engineer time returned modelled	~850 hours a month, 5 engineers' worth
Engineer interruptions avoided modelled	~230 a month

Why it stuck

1. The coding agents use it. 103 people's agents read the same system map their teams do. 2. Answers come from Leena's own code. 53 repositories, 7 million lines, every answer with references into the code. 3. Every team uses it. Engineering, product, QA, design, solution consulting, customer success and support work from the same source. About 230 questions a month that used to land on an engineer no longer do. Security: ProdE has read-only access to the repositories Leena selected. It never writes to them and never trains any model on their code. Indexing is static analysis only. ProdE is ISO 27001 certified.

See it on your own codebase

Connect your repositories to ProdE. Your engineers, your PMs and your coding agents get the same map of the system.

Get a demo