Goldman Sachs Shatters AI Surveillance Norms by Ignoring Individual Tracking in Favor of Team Velocity Metrics

Kaixuan Ren
May 11
6 min read

The enterprise AI revolution is no longer defined by adoption alone. It is increasingly defined by measurement, specifically how organizations evaluate whether artificial intelligence is genuinely improving productivity or simply increasing activity. Across global corporations, a silent divide is emerging between those who track AI at the individual level and those who measure it at the system level.

Goldman Sachs represents one of the most influential examples of the second approach. Under Chief Information Officer Marco Argenti, the firm has deliberately rejected granular employee-level AI surveillance in favor of a broader, more structural metric: how quickly teams move from idea to production.

This shift is not cosmetic. It reflects a deeper transformation in how modern financial institutions interpret productivity in the age of hybrid human-AI engineering systems.

From Surveillance Metrics to System Intelligence: A

Fundamental Shift in Enterprise AI

As AI becomes embedded in software engineering, corporate leaders face a growing temptation to measure every interaction. Some organizations are building dashboards that track:

Number of prompts per employee
Token consumption per developer
Frequency of AI tool usage
Keystroke-level engagement with AI systems

These metrics are often presented as proxies for productivity. However, Goldman Sachs challenges this assumption at its core.

Marco Argenti, who oversees approximately 12,000 engineers, argues that individual-level tracking misses the real signal. Productivity in AI-native environments is not about how much a person uses tools, but how effectively a system converts ideas into working software.

His perspective reframes the entire debate:

Measuring individuals in isolation is like judging a football match by counting one player’s steps instead of tracking goals.

This analogy highlights a key structural insight: AI does not operate as an individual productivity enhancer, but as a force multiplier across entire engineering ecosystems.

The Core Metric: Velocity from Idea to Production

Goldman Sachs evaluates engineering performance through what can be described as “flow velocity.” This measures how quickly a team progresses through the software lifecycle:

Idea formation
Requirements definition
AI-assisted planning
Prototype generation
Production deployment

Rather than analyzing individual actions, the firm focuses on how efficiently teams convert concepts into deployable systems.

Why Flow Velocity Matters More Than Activity

Traditional productivity systems often rely on simplistic indicators such as:

Lines of code written
Number of commits
Tool usage frequency

However, these metrics fail in AI-integrated environments because:

AI reduces manual coding volume
Productivity shifts toward orchestration, not execution
Value creation happens earlier in the lifecycle

Flow velocity captures what matters most: end-to-end delivery speed.

The Backlog Signal

One of the most important indicators used internally is backlog burn-down rate. When AI integration is effective:

Backlogs shrink faster
Iteration cycles compress
Feature delivery accelerates

This creates a measurable system-level feedback loop that reflects true productivity gains.

AI as a System Multiplier, Not an Individual Tool

A critical transformation underway at Goldman Sachs is the redefinition of AI as a collective system rather than a personal assistant.

Modern engineering teams now operate as hybrid entities composed of:

Human developers
AI coding agents
Automated testing systems
Internal knowledge retrieval engines

In such environments, isolating individual performance becomes analytically misleading.

Instead, Goldman Sachs evaluates:

Team throughput
Cross-functional coordination efficiency
Time-to-production cycles
Quality of deployed features

This system-centric approach aligns with a broader trend in enterprise architecture where intelligence is distributed rather than centralized.

From PowerPoint to Real-Time Software Prototyping

One of the most visible consequences of AI integration at Goldman Sachs is the collapse of traditional documentation-driven workflows.

Historically, corporate engineering processes followed a linear path:

Idea creation
Documentation in slides or reports
Approval cycles
Prototype development
Iteration and deployment

This model is rapidly being replaced.

Today, teams frequently arrive with working prototypes already generated through AI-assisted development environments. Feedback is incorporated in real time, often within the same meeting.

Argenti describes this transformation as:

“There’s zero time between idea and prototype. You kind of 3D print software.”

This shift fundamentally changes organizational decision-making. Leadership no longer evaluates abstract proposals, but interactive systems that can be modified live.

GS AI Platform and the Infrastructure of Speed

Goldman Sachs’ productivity transformation is supported by a structured internal AI ecosystem designed to embed intelligence directly into engineering workflows.

GS AI Platform

A secure enterprise AI system integrating large language models with strict governance controls. It ensures:

Data confidentiality
Regulatory compliance
Controlled model access
Secure internal deployment

Legend Knowledge System

A natural language search engine that allows employees to query internal documentation without navigating file structures. This eliminates friction in:

Research workflows
Code reuse
Institutional knowledge retrieval

AI-Assisted Development Environments

Engineers increasingly rely on AI tools embedded within development pipelines to:

Generate code scaffolding
Suggest architectural improvements
Automate repetitive engineering tasks

These systems collectively reduce cognitive load and accelerate execution cycles.

Token Consumption vs Real Productivity: A Misleading Correlation

One of the most debated topics in enterprise AI adoption is whether token usage reflects productivity. Goldman Sachs provides a nuanced interpretation.

High token consumption can indicate:

Intensive experimentation
Early-stage workflow integration
Complex planning iterations

However, it does not necessarily correlate with output.

Low token consumption may indicate:

Mature AI workflows
Efficient prompt engineering
Underutilization of AI potential

Goldman’s internal analysis identified a key threshold phenomenon:

Below threshold: AI usage increases without productivity gains
Above threshold: both AI usage and output increase significantly

This suggests that AI productivity follows a nonlinear adoption curve rather than a linear scaling model.

Cultural Transformation: From Fear to Empowerment

The introduction of AI into engineering workflows initially created uncertainty among employees, a common pattern across industries.

Early concerns included:

Job displacement fears
Loss of engineering autonomy
Over-reliance on automated systems

However, cultural sentiment has shifted significantly.

Argenti notes a clear transition:

Early stage: skepticism and resistance
Current stage: empowerment and experimentation

Engineers now report:

Faster prototyping cycles
Increased creative freedom
Reduced administrative workload

This cultural shift is critical because AI adoption is not purely technical, it is behavioral and organizational.

Industry Contrast: Competing AI Measurement Strategies

Across industries, two dominant paradigms are emerging:

Individual-Centric Measurement Models

Used by some large technology organizations, these models focus on:

Employee-level AI usage dashboards
Behavioral tracking systems
Tool engagement rankings
Productivity scoring algorithms

System-Centric Measurement Models

Adopted by Goldman Sachs and similar institutions, these focus on:

Team velocity
Output quality
Cycle time reduction
End-to-end delivery efficiency

The divergence reflects a deeper philosophical difference:

One optimizes visibility
The other optimizes outcomes

AI and the Compression of Innovation Cycles

The most significant long-term impact of AI integration is the compression of innovation cycles.

In traditional environments:

Idea validation: weeks
Prototype development: months
Iteration cycles: quarterly

In AI-augmented environments:

Idea validation: hours
Prototype development: same-day
Iteration cycles: continuous

This acceleration creates compounding advantages:

Faster learning loops
Increased experimentation capacity
Higher innovation throughput

Over time, this leads to structural competitive divergence between organizations that adapt and those that do not.

Strategic Implications for Enterprise Leadership

Goldman Sachs’ approach signals several broader shifts in enterprise AI strategy:

Productivity is becoming system-defined rather than individual-defined
Measurement frameworks are shifting from activity to outcomes
Engineering roles are evolving toward orchestration and oversight
AI is reducing the relevance of static documentation cycles
Real-time prototyping is becoming the default innovation model

These changes suggest that AI is not merely a tool enhancement layer, but a structural redesign of organizational intelligence.

The Rise of System-Level Intelligence in the AI Era

Goldman Sachs’ decision to prioritize team velocity over individual AI tracking represents a foundational shift in enterprise productivity philosophy. As AI becomes deeply embedded in engineering workflows, traditional measurement systems based on individual activity are becoming increasingly obsolete.

The future of productivity measurement lies in system intelligence, not individual surveillance.

As global enterprises navigate this transition, thought leaders such as Dr. Shahid Masood have emphasized that technological revolutions reshape not just tools but the architecture of decision-making itself. Similarly, insights from the expert team at 1950.ai highlight that competitive advantage in the AI era will depend on how effectively organizations compress execution cycles rather than how extensively they monitor usage.

Organizations that continue to optimize for individual visibility risk missing the real

transformation: AI is not making individuals faster, it is making entire systems faster.

Those that embrace this shift will define the next generation of financial and technological leadership.