In the past few years, one thing has become undeniable: AI is everywhere. Engineers use it to write code. Project Managers (PMs) rely on it to draft specs. Or marketers use it to produce content at scale.

AI is no longer a tool we occasionally use. It’s becoming part of how we work, think, and deliver every day.

And with that, a promise has emerged: “We can now achieve 10x productivity.” In some ways, that’s true. But here’s the uncomfortable truth you need to know.

Productivity Is Growing, But Judgment Might Be Shrinking

Most teams measure productivity through:

Speed
Output
Automation

And yes, AI dramatically improves all three.

But what’s often overlooked is what happens on the other side:

Critical thinking declines
Context gets lost
Expertise becomes shallow
Ownership fades

While they matter more, these are harder to measure. This is what we call a “productivity illusion.”

We feel more productive, but we’re not necessarily making better decisions.

Faster output ≠ Better decisions

The Real Risk: Losing Judgment

One of the biggest risks we see today is “judgment erosion.” When AI workflows are not designed carefully, teams produce more but rely less on their own thinking, and over time, judgment weakens.

It starts when we let AI define the problem, choose the strategy, and make trade-offs. We’re no longer using AI to assist thinking but outsourcing it.

We’ve seen teams go all-in on AI, using it for PRDs, coding, testing, and marketing. At first, the results looked great: things moved faster, and output increased.

But then problems started to show.

Customers said the product didn’t solve their needs
Developers said they built exactly what the spec said.
Product managers confirmed the specs looked correct

Everything was done “right,” but in the wrong direction. The real issue is that human judgment was outsourced to AI somewhere along the way, in defining the problem, strategy, and making decisions.

As a result, speed increased, but clarity was lost.

Rework increased 40%
Team alignment dropped
Skill depth declined

AI x Human: AI Should Amplify Thinking, Not Replace It

So the question becomes: How do we use AI without losing what makes us valuable?

Our answer is simple: Don’t remove humans from the loop. Redesign the loop.

To make this practical, we break AI usage into four layers:

1. Task Automation

Use AI for repetitive, well-defined, low-risk tasks that are easy to review.

Drafting
Formatting
Code scaffolding
Summarization

However, people should still review the results and make the final decisions.

2. Decision Support

Let AI explore:

Options
Possibilities
Trade-offs

But the rule is:

AI suggests. Humans decide.

3. Judgment Layer (Non-negotiable)

This is where humans must stay in control:

Strategy
Context
Trade-offs
Ethics
Experience

Judgment is the last human advantage.

4. Meta Productivity (The Real 10x)

The future is not about using AI but about:

Designing workflows
Setting boundaries
Auditing outputs
Improving systems

Productivity = how well you design the system, not how fast you type.

Better AI x Human Workflow: Human-first, AI-augmented

Many teams operate like this: AI → Output → Deploy (AI-first)

But high-performing teams shift to: Problem → AI assist → Validate → Deploy (Human-first)

Human defines the problem: Clarify requirements, constraints, and what actually needs to be solved

AI generates options: Suggest possible approaches, architectures, or solutions

Human selects direction: Choose the right path based on business context and trade-offs

AI drafts the implementation: Generate code, structure, or first version

Human refines and validates: Review, adjust, and make sure it works in real conditions

This workflow keeps the roles clear:

AI accelerates execution, and human owns decisions.

What Are New Productivity Metrics?

Productivity should not be measured only by output volume.

Traditional metrics like: hours saved, number of completed tasks, speed of execution,etc. are useful but incomplete.

What matters more is:

Time Saved: How many hours can AI and automation help save?
Decision Accuracy: Measures how often the team makes the right decisions. This includes choosing the right solution, solving the correct problem, and making sound products and technical trade-offs
Long-term Impact: Focuses on sustainability and scalability. Questions behind this metric: Will the system scale? Does this reduce future technical debt? Does this create long-term leverage for the business?
Tasks Done: Represents output quantity. However, more completed tasks does not mean better outcomes.
Rework Rate: Measures how often work must be redone. A lower rework rate usually means: clearer thinking, better alignment, stronger architecture decisions, and higher implementation quality
Knowledge Retention: Measures whether the team still understands the system after using AI tools. If developers rely entirely on AI-generated outputs without understanding business logic, architecture, and trade-offs, then productivity may increase while engineering quality declines.

Example: Building With AI Using The 4-layer Model

Let’s take a simple requirement: Create a dashboard that shows user activity, failed logins, and feature usage for admins.

At first, this sounds simple. But building it properly requires going through four layers of human–AI collaboration.

Layer 1: Task Automation handles the mechanical setup

AI generates the initial implementation.

For example, it can create a FastAPI endpoint for /admin/activity-dashboard with fields like:

total logins
failed logins
top-used features

This removes boilerplate work and helps the team start faster.

However, this is only the first draft. The team still needs to validate:

Do the fields match the real PM requirements?
Is authentication handled correctly?
Are we using real data or mock data?
Is the response structure scalable?

AI builds the first draft. Humans own validation.

Layer 2: Decision Support expands the option space

AI does not just write code, it helps the team rapidly compare architectural approaches to fetch the dashboard data. For example, for data sourcing, AI can suggest options and map the trade-offs:

	Option ADirect production query	Option BNightly aggregated table	Option CEvent streaming
Pro	Simple, extremely fast to build	Fast dashboard response, low load on production DB	Highly scalable near real-time analytics
Cons	Heavy queries will impact production database performance	Data is delayed, not real-time	Complex architecture, high infrastructure overhead

However, AI cannot make the final decision because it does not understand:

Business priorities
Infrastructure constraints
Cost considerations

AI can map options, but it cannot decide.

Layer 3: The Judgment Layer anchors the team to reality

This is where humans step in. The team needs to ask:

Are user-level failed logins too sensitive to display?
Do we optimize for speed today, or maintainability tomorrow?
Are there hidden privacy or security liabilities?
Will the simplest solution become technical debt in 6 months?
Does this dashboard really need to be real-time?

These are not technical questions alone. They require context, experience, and responsibility.

This is where real product thinking happens.

Layer 4: Meta-Productivity focuses on designing the workflow

The real value comes from how the team structures the process.

A typical workflow looks like:

Human defines requirements and constraints
AI drafts the initial implementation
Human reviews architecture, security, and data models
AI helps refine and optimize
System automates testing and validation
Human evaluates the final outcome

In this setup, AI handles execution, and humans stay at key decision points

Productivity = better workflow design, not just more code.

Skills That Make Developers 10x More Valuable

To work effectively with AI, developers need to build a different set of skills.

What Developers Should Start Practicing Now?

To accelerate AI x Human collaboration, there are a few habits that make a big difference:

Learn one AI workflow deeply (for example: coding, review, or test generation)
Build a habit of testing AI-generated code first
Get better at reading and understanding existing code
Strengthen system design fundamentals (performance, security, reliability)
Practice explaining technical decisions in simple terms
Set your own rules for when to trust AI and when to review carefully

Conclusion

In the AI era, coding can become cheaper, and execution can become faster. But:

Problem framing
Decision-making
System thinking

Become exponentially more valuable.

At Enosta, this is how we see the shift: We’re not just building products faster. We’re helping teams design systems where AI x humans work together effectively.

If you’re exploring how to integrate AI into your product or workflow, the question is not: “Where can we use AI?” but: “Where must humans stay in control?”

(*) Blog capturing key insights from the talk “Productivity 10x Without Losing Human Judgment” by Mr. Vu Tran, Head of Business Development at Enosta, presented at DevDay 2026.