Productivity Multipliers
Measuring the productivity multiplier of agentic coding vs pre-AI baselines across churn and net code metrics.
Agentic
Coding
Early Findings Benchmark
Part 1: Productivity Multipliers Insights
Marcio Sete | October 2025
There's a widespread scepticism about the productivity multipliers coming from AI coding. I'm experiencing unprecedented productivity whilst my industry peers continue to see only marginal gains.
The Benchmark Setup
Lines of code (LOC) added + deleted in commits merged to main.
Net new code (Adds - deletes) integrated in production, representing codebase growth.
Four Groups
1. Pre-AI Baseline (2020-2023)
Before AI coding was a thing
My control group.
2. AI Augmentation (2024- early 2025)
Where most organisations are now
Using ChatGPT, Copilot, Cursor, and VSCode extensions.
3. Agentic Coding (Aug-Sept 2025)
Soon to be the new normal
One Product Engineer orchestrating a fleet of up to 10 agents working in parallel in one codebase.
4. Founder Mode (Aug-Sept 2025)
The competitive threat
Agentic Coding taken to the extreme: ~100-hour weeks, across multiple product codebases.
The Results
| Baseline Group | AI Augmentation | Agentic Coding | Founder Mode | |
|---|---|---|---|---|
| Churn | 5,986 | 8,079 | 58,853 | 126,499 |
| Net Code | 1,416 | 3,870 | 31,811 | 61,928 |
| Baseline Group | AI Augmentation | Agentic Coding | Founder Mode | |
|---|---|---|---|---|
| Churn | 1X | Multiplier: 1.3X | Multiplier: 9.8X | Multiplier: 21.1X |
| Net Code | 1X | Multiplier: 2.7X | Multiplier: 22.5X | Multiplier: 43.7X |
Key Findings
Agentic Coding delivered ~22× more net new code than the pre-AI baseline.
At the extreme, founders can reach ~44× productivity, churning ~130,000 lines per month and growing product codebases at ~60,000 lines per month.
Net Code/Churn efficiency improved from ~24% (baseline) to ~50% (agentic)-approximately 2× more efficient.
Net/churn efficiency plateaus around 2× as more agents run in parallel-the likely the ceiling of current tech.
Time Compression
| Project Timeline Comparison | |||
|---|---|---|---|
| Pre-AI Baseline | AI Augmentation | Agentic Coding | |
| 12 months | ~4.5 months | 16 days | |
| 24 months | ~9 months | 32 days | |
| 36 months | ~14 months | 48 days | |
| Time to First 100k Lines of Code | |||
|---|---|---|---|
| Pre-AI Baseline | AI Augmentation | Agentic Coding | |
| 1 engineer | ~71 months | ~26 months | ~3 months |
| 5 engineers | ~14 months | ~5 months | |
What This Means for Organisations
Talent shift
One product-engineer with agentic orchestration can approach the output of 20+ pre-AI engineers (order-of-magnitude shift). This completely changes the ratio of engineers, designers, and product managers.
Time-to-market compression
Features that took quarters now take days; products that took years now take months or weeks.
Competitive moats erode
Incumbents' size advantage diminishes when challengers can build 20-40× faster with a tiny fraction of headcount.
The bottleneck shifts
Product strategy, user research, and go-to-market become the constraints, not engineering capacity.
Dopamine-Driven Over-Engagement
In high-velocity "fleet orchestration" workflows, engineers can experience rapid, repeated "wins" (tasks getting completed, tests passing, agents reporting back). This compresses natural recovery cycles that used to occur during coding sessions.
The result is a continuous reward loop-akin to doom-scrolling-where engineers chase "just one more prompt," extending sessions, skipping breaks, and losing somatic cues of fatigue.
Short-term output rises; cognitive cost accumulates. Left unmanaged, this increases the risk of over-engagement, decision fatigue, and eventual burnout, despite nominally higher productivity.
The Incumbent's Dilemma
The question for incumbents isn't "can we work like that?" It's "how do we compete when challengers can?"
Your advantages (capital, customers, brand) matter less if a challenger reaches technical parity before your next board meeting.
What are you going to do about it?
The Question
Can you afford to wait while competitors move up to
22× faster?