Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

3 ways companies can reinvent with AI

If your engineering team uses AI every day, why aren’t you seeing results?

[Photo: Getty Images]

Profile photo of Muamer Cisija

Co-Founder and President · Symphony

Working with private equity-backed technology companies over the past year, we’ve observed a striking paradox. Nearly every software engineering team has embraced AI coding assistants. The tools are installed. The licenses are paid. Daily usage rates hover around 65 percent. By every conventional measure of technology adoption, these companies have succeeded.

Yet when we sit down with chief technical officers (CTOs) and engineering leaders to examine the actual impact on delivery velocity, the story changes dramatically.

Code acceptance rates languish around 27%. More troubling still, 42% of AI-generated code gets discarded entirely—written, reviewed, and ultimately thrown away. Teams are busier than ever with AI, but no faster at shipping software that matters.

This is the adoption paradox we encounter again and again.

The bottom-up fallacy

In our experience helping engineering organizations navigate this transition, the most common mistake is treating AI adoption as something that will organize itself. Leaders purchase enterprise licenses, send around tutorials, and wait for productivity gains to materialize. It seems logical—developers are smart, the tools are intuitive, organic adoption should lead to organic improvement.

“We let it loose hoping it would bubble up,” one enterprise CTO told us recently, reflecting on his company’s initial approach to rolling out AI development tools. “It didn’t.” His team had excellent adoption metrics and disappointing results to show for it.

What we’ve learned from working directly with engineering teams is that unguided adoption creates fragmentation. An engineering VP at a mid-market software company described the situation plainly: “Everyone started using AI their own way. No one standardized.”

Each developer developed their own prompting habits and mental models for when to use AI and when to ignore it. Knowledge remained siloed, and best practices never emerged because there was no mechanism for their spread.

The result is a kind of busy stagnation. Engineers interact with AI constantly—generating suggestions, evaluating outputs, accepting some, rejecting most. Our observations suggest they spend about 10-20% of their time managing this new workflow overhead.

The 27% acceptance rate we observe across organizations tells the story: For every useful line of code AI produces, nearly three get rejected. AI becomes another source of busy work rather than a force multiplier.

Beyond tool adoption

The companies we’ve seen achieve genuine transformation—acceleration factors of two to six times in certain development workflows—approach the challenge differently. They recognize that AI-assisted development requires a fundamental rethinking of how software is built.

As a head of engineering at one of the portfolio companies put it: “It’s not tuning existing processes. It’s reinventing development.”

3 ways companies can reinvent with AI

What does reinvention look like in practice? Three elements consistently distinguish high-performing teams from the rest.

1. Standardization

Successful organizations establish common frameworks for AI interaction—shared prompt libraries, consistent context-setting practices, and agreed-upon workflows for different types of tasks.

This isn’t about establishing a baseline from which individual developers can iterate and improve. When someone discovers a better approach, there’s a mechanism to share it organization-wide rather than keeping it locked in one developer’s head.

2. Proper training

I don’t mean the kind that comes from watching product tutorials. Effective AI-assisted development requires new skills that traditional software engineering education never covered: structured problem decomposition, iterative prompting, quality evaluation of generated code, and knowing when AI accelerates versus when it wastes time.

As one senior developer explained after going through a transformation program, “Problem decomposition has been really key.” Breaking complex tasks into well-defined components that AI can address effectively is a learnable skill, but it doesn’t emerge spontaneously from having access to powerful tools.

3. Mindset shift

This is perhaps most important. The mental model of “developer who occasionally gets AI suggestions” must evolve into something closer to “technical architect who directs AI capabilities.” The developer’s value shifts from typing code to designing systems, from implementation details to architectural decisions, from manual labor to quality control and strategic oversight.

What transformation looks like

When the pieces come together, the change is unmistakable. We recently worked with a senior developer who, after several months of structured AI integration, offered this assessment of his daily work: “I don’t write code anymore.”

His role had transformed from producing lines of code to orchestrating systems, including:

  • Defining requirements with precision
  • Designing architectures that AI could implement effectively
  • Reviewing and refining AI output against rigorous quality standards
  • Ensuring security and maintainability.

He was delivering substantially more value than ever before, just in a fundamentally different way.

This is the gap between the roughly 5% of engineering organizations that have genuinely become AI-first and the vast majority still stuck in high-adoption, low-value mode.

The former have reimagined what their developers do and how they create value. The latter have simply given their developers another tool to add to an already crowded toolkit.

The path forward

Right now, AI-assisted software development feels transitional because the tools have matured faster than the organizational capabilities to leverage them. Most companies have cleared the easy hurdle—making AI available—but few have tackled the harder one of making AI transformative.

Substantial and durable competitive advantages will accrue to companies that have done the harder organizational and cultural work of reimagining how their teams operate. They’ll be the ones where senior developers say things like “I don’t write code anymore”—and mean it as a testament to how much more valuable they’ve become.

The gap between AI adoption and AI value capture is real, but impermanent. The question for every engineering organization is whether they’ll close it themselves—or watch their competitors do it first.

Muamer Cisija is the cofounder and president of Symphony.