A little more than three years after the launch of the ChatGPT that heralded an era of generative AI promise, companies are finding the going hard. This is the demanding phase of technology adoption. Organisations are looking to scale, seeking effective ways of turning artificial intelligence pilots and proofs of concept into measurable and meaningful applications. The transition isn’t always simple. As a result, these firms are looking to answer two questions – where next? And how do they get there?
To cast an eye on the journey and the path to it, Tech Monitor and AMD brought senior IT professionals together for an executive roundtable in Sweden’s capital, Stockholm. ‘From AI promise to business impact: building future-ready enterprise AI’ took place in mid-March.
On good data, one attendee diagnosed the problem. He said organisations often lacked a unified data platform while, simultaneously, showing a tendency to “pull data away from its source”. Another said too many organisations didn’t fully know where their data resides or whether it can be trusted. “Without good data, AI is nothing,” he said.
To overcome the data problem, another voice floated the idea of using synthetic data – cleansed and stripped of its biases. On the surface, this would appear to be a positive pursuit, ensuring that AI models are not poisoned by prejudices and predispositions. Most agreed but more than one attendee pointed out that if you wanted to understand customers behaviour – or citizen behaviour in a public service context – you need to find a way too “absorb the biases”. On a similar theme, another voice around the table noted that his organisation is experimenting with “noise injection” to ensure their AI models are robust.
On good architecture, more than one attendee noted how silos and legacy systems continue to characterize the modern enterprise. Resolving enterprise complexity has been a long-term objective and AI might finally provide the impetus. “We can’t postpone it any longer,” said one voice around the table, arguing that AI’s advance is at risk if we don’t.
Other reasons for slow progress included unrealistic “expectation setting” with more than one senior voice urging that senior management get better at distinguishing between hype and reality. To this end, IT professionals need to become expert at communicating upwards. One way to achieve this is to ensure that at least some of those in leadership positions have better technical knowledge so they can assess whether a proposed project is viable. “Spreading AI awareness is one of the biggest responsibilities of senior management,” insisted another voice around the table. “We need,” he said, “to be able to explain what the art of the possible looks like, and what is not possible.”
If GenAI progress is slowing in some parts of the economy, there is evidence that elsewhere organisations are putting Agentic AI into action. One such company has implemented four Agentic AI applications after an initial workshop spawned 40 candidate use cases. From there, the organisation identified a handful to operationalise choosing those that were least complicated and demonstrated fewest variations but still provided tangible benefits. All are backend applications and all play to operational efficiency.
One is a case management system that can distil multiple information sources – emails, PDFs and more – in order to address customer queries. Another is an HR agent that helps answer employee requests in an organisation with more than 16,000 employee agreements. Another company represented around the table has a similar application – colloquially referred to internally as the “mind reader” – that can tackle the most common human resources-related queries. It answers the 80 percent, leaving the 20 percent that are more complicated to human intervention.
During the discussion, agents were described as “team members”. Today, their boss is human but that needn’t be the case long term. These artificial employees – typically “junior members of the team” at first – are likely to be “promoted” as they learn and improve. Asked whether this anthropomorphising of agents might lead to resentment among the human workforce, this proponent of agentic AI insisted that there was “plenty of excitement” within his organisation. Besides, he added, agents are taking care of “the dull work” the human workforce is reluctant to tackle.
This small sample might suggest that agentic AI is in full flow but many around the table are in “listen and learn mode” not yet ready to embark on their own projects. Indeed, another attendee offered a counter view, suggesting 2026 may prove to be “a year of failure” as more organisations test out agentic ideas.
But as always with technology adoption, failure is not an end point. Rather it is a means of learning lessons and discovering potential paths to success.
‘From AI promise to business impact: building future-ready enterprise AI’ – a Tech Monitor / AMD executive roundtable discussion – took place on Wednesday 18 March 2026 at Villa Dagmar, Stockholm, Sweden.
