Why Complex Software (and AI) Projects Keep Failing: And What Actually Works

Despite massive investments and decades of experience, most complex software and AI projects still fail, and the reasons may surprise you. Companies pour millions into new technology, hire expert teams, and adopt the latest methodologies, yet the majority of initiatives never deliver the promised results. The problem isn’t new, and the lessons from decades of project tracking remain painfully relevant.

Since 1994, The Standish Group published the Chaos Report, tracking software project success rates year after year. The results were consistently sobering: only a small fraction of projects truly succeeded. The last report indicated a success rate of just 31%. A successful project was defined as one that delivered on time, on budget, and met the agreed-upon feature scope.

In 2020, The Standish Group stopped publishing the report, arguing that software could no longer be treated as a one-off project to be managed in the traditional sense, but required continuous and staged delivery as business needs kept changing.

I wish the Standish Group had continued the report, because the underlying problem remains, and the situation hasn’t improved.

A recent MIT report found that 95% of AI projects fail. Surprisingly, the reasons at a macro level echo those identified decades ago:

• Clear business objectives

• Executive support

• Emotional maturity

• User involvement

• Process optimization as the focus

• Skilled personnel

• Strong software production, deployment, and operations practices

• Effective methodologies (e.g., Agile)

• Project management expertise

  
  

While all of these factors matter, I’d point to a key culprit that is often overlooked: code complexity or technical debt. The more complex the code, the more complex the project. The relationship is exponential. Ten times more code often means a hundred times more complexity to maintain.

Why has SaaS been successful? Large monolithic solutions often failed because of excessive complexity. Executing in smaller, modular pieces proved far more manageable. Similarly, AI agents succeed at small tasks in specific environments because control and management remain feasible. As systems grow, the combinatorial complexity skyrockets.

With vibe coding (AI-driven code generation), the amount of code does not decrease. In fact, it does nothing to ease maintenance and adaptation and may even make it more difficult.

That’s why I continue to trust low code platforms, with vibe coding included as an optional add-on rather than the backbone. Architected products that flex to different user needs allow innovation to accelerate without being hampered by technical debt. With low code, 90% of the code does not need to be maintained, and that’s a huge advantage.

Even before organizations embark on agentic AI, most have large gaps in basic process automation. A very large proportion of organizational processes are still run manually or with spreadsheets. Basic process automation is a prerequisite for AI to truly deliver.

We’ve been on the low-code journey for over 15 years. The Stragiliti Low Code platform (www.Stragiliti.net) has matured to the point where complex applications can be built with very little code. Our success rates for projects are very high. In fact, we harp on project certainty in the projects we execute.

For portions that do require coding, we now use AI to accelerate and enhance outcomes even beyond the typical low code promise of 3x to 10x lower effort. Low code will continue to remain our backbone, ensuring that technical debt never undermines progress, allowing us to focus on all other aspects of project success with ease.

Trust us with your next enterprise custom application project and experience the Stragiliti low code + AI difference in terms of speed, quality, affordability and most importantly success certainty.