Why Your Job isn't at Risk When You Own the Value Creation

Partha Mishra's career spans over two decades, beginning in the mainframe era when developers worked with raw algorithms and fewer frameworks. As Encora's Data and Analytics Practice Leader, he has navigated the peaks of distributed data processing, RPA automation, and now the AI revolution.

Through every shift, one bottleneck remained: research and evaluation. Finding the right algorithm, evaluating which worked best—developers spent weeks, sometimes months on this. What strikes Partha about AI is that this research capability is now automated. That changes everything. But beneath the productivity metrics and job anxiety, he sees a different reality: the professionals at risk aren't the ones using AI tools. They're the ones who haven't figured out that synthesis, judgment, and decision-making remain entirely human.

After more than two decades in development, what feels fundamentally different about this AI shift?

AI is essentially that block of research becoming available through automation rather than through people. The biggest challenge we always faced was optimization. Finding the correct algorithm and evaluating which one worked best took an enormous amount of time. Now, AI provides automated research capability.

Everyone talks about productivity gains. What are you actually seeing?

Business analysts used to take three months to arrive at meaningful requirements. Now they start at month one. That's significant time savings. Designers get 30 to 40% time savings in the initial phases. Developers can create a first version in one or two days. The initial starting phase sees 50 to 60% acceleration.

We built an accelerator called EnMapper for data migration and warehouse creation. It's backed by LLMs and our AI platform, AIVA. What used to take three months for source-to-target mapping now takes one month. It automatically profiles data, does domain-based field segregation, and creates meaningful mappings. That's a real example of repeatable productivity gains.

That speed changes what developers do. How would you describe the shift?

Before it was "How do I write it myself?" Now it's "How do I use this tool effectively?" Memorizing syntax became less critical. What matters now is knowing how to communicate with AI tools and orchestrate which tool to use for what purpose. You start as a reviewer rather than a fresh developer. Someone has already written the initial code for you.

Is this approach the future of development or just useful for prototypes?

This isn't entirely new. We've always had automation helpers. What's new is the scale. Instead of knowing which function to use, you get entire working code sections. AI learns from massive amounts of code and can even learn from your company's codebase.

But three skills have become critical. First, be very specific about what you want. Without clear instructions, AI might create complicated solutions. Second, get good at reviewing code. Developers love building from scratch but often dislike reviewing what someone else wrote. You need to practice reviewing carefully. When problems show up later, you'll fix them yourself. Third, test thoroughly. Testing ensures that everything works correctly and prevents the addition of unnecessary code.

You've mentioned seeing AI-generated code pass all tests but fail catastrophically in production. What happened?

Here's a real example. There's a technology called Spark that evolved from an older approach called RDD to a newer, faster one called DataFrame. Both patterns exist in AI training data. One developer uses AI with the newer approach, and another uses AI that generates the older style. When you combine these, the system converts between them for every record, resulting in massive performance issues.

Functional testing won't catch it. The code does what it's supposed to. It only shows up in load testing. A good QA must now watch for transitions between different AI-generated code patterns. While AI can accelerate functional testing, catching these mismatches requires more human attention than before.

With all this complexity, where does human value live now?

Think of AI as a tool rather than a replacement. Take a business analyst. AI can do the initial research and drafting. The BA starts day one with month one's work complete. But the next phase requires human expertise: bringing multiple AI tools together, choosing the best outputs based on experience, synthesizing outputs into something coherent.

If AI does the synthesis, that's AI work, not BA work. Stakeholders want the BA's judgment, prioritization, and decision-making. There are three phases of value creation: learning the tool, mastering the tool, and creating value through synthesis and decision-making based on experience.

At no point is the job at stake. The professional creates the output and is responsible for it. As long as you're clear about the value you create, the credit goes to you.

For people entering development now, what makes someone valuable?

Two things. First, explore different tools. Don't get locked into one. Getting exposed to various tools is the first step toward learning synthesis. The variations make us rich. You need to understand and respect the differences.

Second, understand the basics. Take a step back to understand how these tools work. Return to classical machine learning, understanding neural networks, CNNs, RNNs, and transformers. When the mystery behind the magic becomes clear, you gain confidence. The tools have the power to sway decisions, but confidence comes from understanding what's happening behind the scenes.

Partha's focus on quality checkpoints and human synthesis reveals critical technical realities of AI development. To understand how this plays out at the organizational level and what leaders need to know about navigating this shift, explore Sonali Kochar's perspective in From Writing Code to Solving Problems. Or return to the full exploration: Unlike Anything Before: How AI Has Transformed Software Development.