Every now and then you attend a talk that doesn’t present something radically new, but instead puts into words, with unusual clarity, what many practitioners are already living day to day. That was my experience listening to Seun Ebenezer (CDMP, BCS, DAMA) ’s recent lecture at the DATA NEXT – Trust Europe Summit (thank you, Philip Bell) on the future of Data Governance, Data Quality and Data Integrity.
This was not a speculative session about what might happen “one day”. It was a pragmatic story about what breaks when organisations scale faster than their governance models, and how GenAI agents can actually help if they are introduced with care, proper sequencing, and human accountability at the centre.
Greencore is not a theoretical case study. With more than 30,000 employees, over 40 manufacturing sites, and a business scaling from £1.5B towards £4B in annual turnover, the company operates where data velocity, regulatory pressure, and operational intricacy collide on a daily basis. Especially in a highly regulated food industry, governance now needs to operate almost in real time.
And that is precisely the point at which traditional, human-only stewardship models start to struggle.
When Governance can’t keep up
The challenge Seun Ebenezer (CDMP, BCS, DAMA) described will sound familiar to many teams. Service backlogs grow as data requests and remediation tickets pile up. Metadata goes stale because no one can update definitions, lineage, and classifications fast enough. Data stewards, often doing the role part-time on top of their day jobs, end up buried in admin instead of focusing on decisions, exceptions, and real risk.
None of this is a failure of intent or capability. It is a mismatch between operating models designed for a slower data reality and the exponential growth of modern data ecosystems.
As Seun put it, governance has to become as adaptive as the data it is meant to govern.
Meet “Bob”: GenAI as stewardship capacity, not authority
The most compelling idea in the lecture was not automation for its own sake, but the introduction of GenAI agents as extra hands inside the stewardship layer. Seun affectionately named this agent “Bob”, and the name stuck for a reason. Bob is not a replacement for people. Bob is capacity.
In Greencore’s proposed model, GenAI agents step in to handle the work that humans simply cannot scale:
Continuously scanning catalogues to detect outdated or inconsistent definitions across ERP and operational systems.
Profiling data quality, spotting patterns and anomalies, and flagging issues long before manual processes would notice.
Surfacing privacy and PII risks by checking datasets against policy constraints.
Accelerating lineage mapping, metadata discovery, and backlog triage.
What matters just as much is what Bob does not do.
Agents do not own risk. They do not make final decisions. They do not carry accountability. Humans do… always.
That distinction is not cosmetic; it is foundational.
You can’t build the house from the roof
As we say in Spanish, no se puede empezar la casa por el tejado — you can’t start building a house from the roof.
GenAI-assisted governance only works if the underlying data is already fit for purpose. Without high-quality, well-governed, AI-ready data, there is no safe or scalable AI.
No data, no AI.
Seun Ebenezer (CDMP, BCS, DAMA) was very explicit about this sequencing, and it is one of the most important messages to take away. Before introducing agents into stewardship, organisations must do the unglamorous work: tightening catalogues, stabilising business glossaries, mapping lineage, defining ownership, and agreeing on approval paths. AI does not fix weak data foundations, it amplifies them.
Only when that foundation exists does GenAI become an accelerator rather than a liability.
A human–AI operating model, not a replacement
What emerges from this approach is not a break with established governance principles, but an evolution of them. The traditional DAMA view was designed around human roles because, until recently, humans were the only realistic option. That assumption no longer holds.
In the hybrid model Seun outlined, GenAI stewards take on many operational governance tasks, from technical metadata discovery and glossary maintenance to DPIA/ROPA scoping and continuous data-quality monitoring. Human stewards, in turn, shift their focus towards review, approval, escalation, and judgment.
It’s a subtle but powerful shift. It frees expert stewards from administrative drag and lets them spend their time where it actually matters: applying context, exercising accountability, and navigating ethical trade-offs.
Even governance councils evolve: agents participate as signal generators rather than decision-makers, enabling faster detection and remediation while preserving human authority at the top of the pyramid.
Human-in-the-loop is not a constraint, it’s a superpower
A lot has been written about human-in-the-loop oversight, usually framed as a regulatory checkbox. In practice, it is an operational advantage.
The EU AI Act makes this clear, particularly in Article 14 on human oversight and Article 9 on continuous risk management, but regulation is really just catching up with what good practitioners already know. Oversight only works if humans can audit behaviour, understand patterns, and trace decisions back to data and process.
Many of us have been doing this for years through root-cause analysis, quality investigations, and post-incident reviews. GenAI does not remove the need for this work; it multiplies it. To audit an agent’s behaviour, you need a clear view of how data flows across organisational layers, how systems communicate, and how semantics are shared.
A “single source of truth” does not mean forcing everyone to work the same way. It means agreeing on what the data means, even when teams interact with it differently.
Once that shared understanding exists, organisations can move from reactive governance, fixing issues after the fact, to proactive governance that anticipates risk and enables better decisions.
Before we build Bob 2.0
What makes Seun’s vision so compelling is not the technology itself, but the respect it shows for people. GenAI agents are introduced not to sideline stewards, but to recognise their value — to give them back the time and cognitive space needed to think, decide, and lead.
There is nothing more human than that.
Before we rush to build “Bob 2.0”, we should make sure we are building on solid data foundations, shared semantics, and clear accountability. Only then can GenAI genuinely help governance scale — not by taking humans out of the loop, but by putting them exactly where they belong.