For the Enterprise, Coding Co-pilots Are Simply Not Enough

The hype around one specific part of next generation software development and delivery, AI-powered coding copilots, has reached new heights. Over the past few years, we have seen offerings from traditional DevOps vendors like GitHub (GitHub Copilot), newer entrants like Cursor, and even the foundational model companies like OpenAI (Codex), Anthropic (Claude Code), and more. These tools all promise faster code creation, the removal of repetitive tasks, and an economically sound approach to prior human-based pair programming approaches. At the same time, these tools are being positioned to leaders of large-scale development organizations as a means to improve software development economics, drive innovation velocity, and ultimately improve the business process of software development and delivery.

But here’s the early and uncomfortable truth for enterprises: coding copilots rarely address the biggest bottlenecks and are not moving the needle on business outcomes as promised.

I have zero doubt that coding copilots are here to stay. They can and do accelerate local developer productivity. But the software development and delivery process (especially in large scale enterprises) is a complex and interconnected system spanning planning, coding, testing, securing and releasing applications into production, all while maintaining proper governance and compliance. Copilots alone only improve coding efficiency and miss the bigger opportunity: improving flow, security, and quality across the entire software lifecycle.

The Shocking Limitations of Coding Copilots in the Enterprise

Coding copilots have been adopted faster than any other AI solution within development organizations. Recent estimates indicate in the last two years alone over 90% of enterprise R&D organizations have either fully adopted or piloted copilots. And yet the impact of both the change management and tooling cost have been mixed. Two specific reports in recent months have sent shockwaves of disappointment throughout the industry, including METRs findings that showed copilot adoption has actually made developers slower and the now infamous MIT study that showed 95% of all AI projects in the enterprise have “failed”.

The big question is – why are coding copilots not having the expected impact in the enterprise? The failures span various simple realities:

• Code generation is just one step in a bigger process – Coding copilots are visualized and engaged within an IDE (integrated developer environment). They are world class at suggesting code snippets, design patterns, and boilerplate code, but they are often blind to the broader context especially in larger, more complex environments. They lack knowledge of business priorities, architectural standards, security requirements, and compliance rules
• Coding copilots amplify bad planning practices upstream – In most enterprise companies the planning process is much more laborious and time consuming than coding. Prioritization, work break down efforts, and task assignment is often measured in months not days or weeks. In fact, many of the customers we work with every day spend 5-10x more time in planning than coding. To make matters even more challenging, if upstream planning is flawed (unclear requirements, misaligned priorities, disconnected roadmaps), copilots simply help developers build the wrong things more quickly. Speeding up and automating misdirection doesn’t create value, it compounds waste.
• Integration and delivery bottlenecks downstream – Code, whether created by a human or a machine, needs to be tested, secured, scanned, and delivered. If downstream processes are slow, manual, brittle, or fragmented, any gains in coding time are unlikely to translate to faster or more effective delivery. Coding copilots in the enterprise often do not address the true bottlenecks that exist downstream and thus impact is muted.
• Enterprise scale and complexity – There is an old saying in software development “code is read 10x more than it is written”. This is even more true in the enterprise. Unlike startups, large enterprises wrestle with legacy systems, complex architecture, massive code bases, globally distributed teams, and strict regulatory realities. Coding copilots don’t understand these challenges and thus do not address them.
• The Math Doesn’t Math – As the name would suggest, coding copilots target developers. It turns out, on average, only 50% of the people in an enterprise development organization are actually developers. They are joined by designers, architects, QA professionals, etc. To make impact even more challenging, the 50%, on average, spend only 25% of their time writing code. Often, they are in meetings, doing research, or whiteboarding new ideas. With the most positive early reviews of copilots showing a 10-30% developer gain – the maximum impact today is 50% x 25% x 20%, which equals a 2.5% of maximum improvement on the overall process.

The Bigger Unlocks Lie Upstream and Downstream from Coding

This goes to the heart of why we exist. By design, Digital.ai exists to improve and optimize the business process of building and delivering software. So while we believe in coding copilots, our data and our customers tell us the real flow unlocks happen when we improve the automation and connective tissue before and after coding.

Upstream: Agentic Planning

Enterprise companies spend up to 50% of the total R&D time in planning. Leveraging AI to drive a more agentic approach to planning speeds time from idea to development while also improving decision making to avoid last-minute change and unexpected conflicts. When planning becomes more intelligent and adaptive, the impact of coding copilots delivers increased innovation and improved alignment with business objectives.

Downstream: Agentic Testing, Agentic Security, and Agentic Delivery

Adopting Agentic Planning upstream of coding is a major unlock—but the biggest opportunities lie downstream. Advances in Agentic Testing are helping ensure software quality across an ever-expanding range of devices and environments. Agentic Security is accelerating delivery while strengthening defenses, enabling apps to be hardened early in development and intelligently protected in production. And smarter delivery pipelines—blending agentic, automated, and human tasks—are speeding value delivery without sacrificing control, compliance, or governance. Together, these downstream innovations turn raw code into business value faster, safer, and with less friction.

Smarter Delivery Over Coding

AI is powering a renaissance in the world of software development and delivery. Every organization needs to be thinking differently in this 4th Wave. Real productivity gains come not from improving and optimizing isolated tasks like coding, but from removing friction in the end-to-end flow of work. Coding copilots are part of that story, but they are not the ultimate headline.

When enterprises invest and innovate upstream (agentic planning) and downstream (agentic testing, security, and delivery), they can unlock the true exponential gains promised in this 4th wave of software development: more business value, faster time to market, reduced risk, increased security, and more predictable outcomes.

The focus of my next blog will be prescriptive “How to” roadmap for enterprises that are ready to unlock 4th Wave gains.