The Rise of the Lobster: Smarter AI, Same Failures
Agentic AI might be new. The way organisations are approaching it isn’t. In this article, Martin Adams breaks down why the same issues that derailed RPA are showing up again, and how to avoid repeating them.
Read on LinkedInNo, this isn’t the title of a bizarre new film. It’s an emerging colloquial term for agentic AI and the growing excitement around it.
I recently attended the CDAO Sydney Conference. Unsurprisingly, AI dominated the agenda. What did surprise me was how familiar it all sounded.
Glossy presentations. Eye-catching statistics on adoption. Equally eye-catching statistics on failure.
A strong sense of déjà vu.
I’ve seen this before.
RPA.
You could argue RPA was the predecessor to agentic AI. It automated tasks based on rules. Agentic AI embeds decision-making intelligence from large language models, such as ChatGPT, into those processes.
But here’s the key point.
Adding intelligence doesn’t remove the fundamental challenges of automation.
Applying new technology to the wrong process still fails. Poor data still produces poor outcomes. And automation still doesn’t fix broken foundations.
Conservatively around 42% of AI initiatives are abandoned (CIO Dive, 2025). RPA saw failure rates of 30–50% (EY, 2017–2020).
Different technology. Same pattern.
So how do we break this cycle?
AI absolutely has the potential to transform how we work. But success requires a balance between strategy and experimentation.
Lean too far into technology and you end up forcing use cases that don’t matter or building solutions that don’t scale or can’t be secured.
Lean too far into strategy and you move too slowly, while others leapfrog you. With AI, that risk is real.
The organisations that get this right do both.
A simple approach:
Identify multiple high-value processes across the business. Prioritise them based on impact, not novelty.
Run open ideation. Combine those excited by new technology with those who challenge value, risk and security.
Assess the foundations. Data quality, process maturity, integration – if these aren’t right, the outcome won’t be either.
Factor the cost of fixing foundations into your prioritisation. Not every opportunity is worth it.
Understand the human impact. Take people on the journey. Show how they benefit, not just what changes.
Validate demand. If the teams involved aren’t bought in, either the story hasn’t landed or the problem isn’t real.
