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.

Everyone has an AI strategy except the people making it work

Executives set AI direction. Engineers experiment with tools. But the layer in between—the one turning strategy into daily engineering decisions—has been left to figure it out alone.

[Photo: Getty Images]

Profile photo of Vignesh Durai

Software Engineering Manager · Sentry Insurance

At a developer conference earlier this year, I ended up at a dinner table with six engineering managers from different companies. None of them knew each other. Within 10 minutes, every single one was describing the same problem.

Their executives had announced AI strategies. Their engineers were excited to use new tools. And they—the managers—were the ones expected to make it work. Without new resources, definite policies, or anyone acknowledging that the job had fundamentally changed.

One of them said: “Leadership says we’re an AI-first company now. My team asks me what that means on Monday morning. And I’m supposed to have the answer.”

That conversation has stayed with me because it wasn’t unique and it’s one I continue to hear. Organizations are building AI strategies at the top and buying AI tools at the bottom. The middle—the engineering management layer that translates strategy into practice—is being left to improvise.

THE GAP BETWEEN STRATEGY AND SPRINT

Every AI strategy I have seen follows a similar arc. An executive team identifies AI as a priority. Tools get procured or pilots get launched. And then the strategy reaches the engineering floor, where someone has to answer questions that the strategy document never addressed.

Questions like:

  • Can engineers use AI-generated code in production, or only for prototyping?
  • Who reviews that code—and to what standard?
  • How do we estimate work when a model drafts the implementation in minutes but the review takes three times longer?

These are not theoretical questions. They come up every week, and the person fielding them is the engineering manager—not because the role was designed for it, but because no one else is close enough to both the strategy and the code. The daily work of making it real is happening through informal decisions made by managers who were never given the mandate or the time.

AN INDUSTRY PATTERN, NOT AN INDIVIDUAL PROBLEM

This is not a problem at one company or in one industry. It is a structural pattern across enterprise engineering.

I have talked with engineering managers at healthcare companies navigating AI compliance questions their legal teams haven’t answered yet. Managers at financial services firms are told to adopt AI responsibly without guidelines on what that means. Managers at mid-size SaaS companies who ran AI tool pilots on their own initiative, produced results, and then were asked to scale them—still with no additional support.

The translation work—the hard, unglamorous work of making AI adoption safe and effective falls to the engineering manager.

That work includes things most executives don’t see such as:

  • Writing informal AI usage guidelines because formal ones don’t exist.
  • Redesigning code review processes to handle machine-generated pull requests.
  • Coaching engineers through the emotional side of the transition.
  • Performing all these tasks while still being measured on the same metrics as before: shipping velocity, team retention, and sprint predictability.

WHY THIS IS A STRATEGIC RISK

When the translation layer is unsupported, AI adoption becomes inconsistent. One team adopts tools thoughtfully because their manager invested personal time in designing a process. The team next door adopts haphazardly because their manager was too overloaded. Both teams report AI adoption to leadership. The outcomes are wildly different.

There is also a retention risk. When the scope of the job expands without acknowledgment, the best managers quietly start looking for organizations that recognize what they are doing.

And there is a quality risk. AI governance decisions made informally by overloaded managers are fragile. When something goes wrong with AI-generated code in production, the question will be: who approved this? The answer is often a manager who was never formally given that authority.

3 THINGS ORGANIZATIONS CAN DO NOW

Closing this gap is less about budget and more about clarity. Three changes would make an immediate difference.

1. Close the policy gap before the tool gap. If you’re buying AI tools for engineering teams, publish use guidelines. Identify what code is acceptable for production, the data that can be shared with external models, and the decisions that need organizational ownership.

2. Make the translation work visible. Name, resource, and measure the expectations for engineering managers to frame how they evaluate tools, design AI-augmented workflows, and guide their teams through the transition. Performance conversations should reflect the new job scope.

3. Create a peer network for managers navigating AI adoption. A lightweight community of practice—internal or cross-company—where managers share what is working and what is failing is more valuable than most top-down training programs.

STRATEGY WITHOUT TRANSLATION IS JUST A SLIDE DECK

After that dinner, I stayed in touch with a few managers from the table. One recently started a small working group for engineering managers at her company—a space to share AI governance decisions and lessons learned. No executive mandate or budget. Just managers helping each other close the gap.

She said something that remained with me: “We are all building the same bridge. We just didn’t know anyone else was building it too.”

That bridge—between AI strategy and engineering reality—is where transformation happens. Not in the boardroom or the model. In the daily decisions made by people who turn ambition into working software.

The question is not whether your organization has an AI strategy. It probably does. The question is whether the people responsible for making it real have what they need to succeed.

Vignesh Durai is the software engineering manager at Sentry Insurance.