According to a recent widely reported MIT study, generative AI projects have a stunningly poor success rate. I have a few thoughts on this.

Any study is just one data point, and this one is likely addressing very large-scale projects. But it's an important indicator, demonstrating something that's easy to forget in the face of this amazing new technology: even though AI is different than previous technologies, the fundamentals of technology adoption still apply.

In other words: AI is new, but its adoption needs to follow the exact same playbook that we all know and love. Namely, aligning with business needs, designing user experience, overcoming resistance to change, training, support, feedback loops, testing, project management, communication, etc.

Consider some of the points raised in that report:

• Model quality wasn't an important factor, but its integration into workflows was.
• Success is markedly higher when purchasing from third-party vendors.
• Successful pilots tended to focus on specific pain points, in particular back-office automation rather than sales and marketing.

What lessons can we learn from this?

• It's not about the technology, it's about how you use it.
• Complexity kills.
• Avoid shiny-object syndrome.
• Don't reinvent the wheel.
• The closer you are to your end-users, the better feedback you'll receive.
• Stay focused.

All of these are time-tested and well-known strategies.

There's one more lesson that I cannot emphasize strongly enough. It wouldn't surprise me if this underlies many of those 95% failures. First, observe from the report:

• Adoption of AI isn't leading to layoffs, but it is leading to slower backfill/re-hiring when people leave.

In practice this means AI isn't replacing workflows, it's augmenting them. From that I draw my primary advice for designing user experience in the age of AI:

• AI doesn't replace the human, it assists the human.

Focusing on assistance-based workflows greatly simplifies deciding how best to leverage AI. Ironically, that same principle applies when adopting almost any new technology. It's as if AI is just like any other technology...