The dramatic progress of AI and its impact on both the economy and society at large dominated the conversation throughout the World Economic Forum at the Swiss resort of Davos.
For many, this will have prompted the view that enterprises are on the precipice of epochal change, through either the advent of artificial general intelligence (AGI) or AI sweeping away the old ways of how enterprise software is designed, built, and run.
But let’s take a step back for a moment. For all the talk about AGI at Davos, the short timelines for its arrival that were floated in public felt, at best, fanciful. Sessions that leaned into these speculative futures were highly visible, yet what was discussed in the rooms and hallways nearby was very different.
For business leaders attending the event, AGI was a footnote.
Business leaders frustrated over AI projects that fail to deliver
The real conversation, grounded in their early experiences with AI, revolved around a much more immediate challenge: how to use AI at scale with systems that were never designed for it.
The huge obstacle regularly encountered is that they are integrating AI into workflows that are still designed with the expectation that a human will do the work – not automation or AI.
The result is a mess, and there was no shortage of frustration about big AI projects that failed to deliver more than a good-looking demo.
There must now be a thorough rethinking of how AI is used to develop enterprise applications at scale. Certainly, louder voices than mine addressed this subject too.
NVIDIA’s Jensen Huang unveiled his vision of how software would be built on AI. That grabbed my attention, not least in how he talked about a five-layer cake model. That analogy to ‘layers’ struck a chord.
I agree that AI is the future of software but there’s some devil in the detail. You need the right ingredients to fold into any layer cake recipe to arrive at AI-boosted enterprise software that’s powerful, predictable, and delivers the correct business outcomes.
To us, you need a ‘workflow’ layer that helps govern the AI outcomes by providing repeatable and validated pathways for the parts of the business where it makes sense to have figured these things out in advance.
The opportunity is to rethink how applications are conceived at design time, so AI can deliver predictable value at run-time.
When agentic AI is used at run‑time, it must be deeply grounded in the context of the available workflows and predictable tools. These include business rules, deterministic automation, and structured data sources that can be orchestrated together, instead of relying solely on free‑roaming LLM-based agents that can go rogue.
AI failures are strategy failures
From what I heard privately at Davos, AI failures are largely strategy failures. The ambition and excitement to use AI to transform an organisation has run aground on ill-defined outcomes, broken processes, and technical debt. Business leaders are beginning to recognise this.
The next step is to harness the power of AI into practical answers and solutions that actually clean up the mess. From there, the ultimate objective, and the real challenge for all of us, is to move the conversation from forward-thinking rhetoric to the creation of tangible, sustainable, long-term transformation.