On a warm June evening in Shoreditch, engineering leaders from across industries gathered for another EngineeringX dinner to discuss one of the biggest questions facing technology organisations today:

How do we preserve the speed and innovation AI enables while maintaining trust, governance, accountability, and commercial discipline?

The conversation started where most AI conversations start: coding assistants, agents, automation, and productivity gains. Fortunately, it became boring. The “same old”  very quickly didn't stay there for long.

What stood out throughout the evening was how honest the room was. There was very little AI hype and a healthy amount of sarcasm. Many of the leaders around the table had already moved beyond the excitement phase and were now confronting the harder reality of what AI adoption actually means inside small and large organisations.

There was also a noticeable frustration with the growing sense of AI FOMO. Several guests described boardrooms and executive teams feeling pressure to "do something with AI" before fully understanding what organisational changes would be required to make that successful.

The conversation kept returning to the same tension:

AI is not primarily a technology challenge. It's an organisational design challenge,

And many organisations are still trying to solve tomorrow's problems using yesterday's operating models.

AI Didn't Create a Governance Problem. It Exposed One.

One observation surfaced repeatedly throughout the discussion:

"AI hasn't created a governance problem. It simply exposed the governance problems we already had."

Many guests reflected on previous transformation waves, particularly cloud adoption.

One leader remarked that many organisations spent years moving infrastructure into the cloud while leaving their governance models largely untouched.

"We migrated the servers, but not the thinking."

The room recognised the pattern immediately.

The challenge was never really infrastructure. It was operating models.

AI is now exposing exactly the same issue, only faster.

Several guests argued that traditional governance mechanisms were already struggling long before AI arrived. Human approval gates, change advisory boards, manual checkpoints, and compliance processes built around increasingly large checklists were becoming difficult to sustain in modern delivery environments. We've built, and are still building, governance nuclear power plants; layers upon layers of oversight designed to prevent any single failure. It doesn't have to be that way with AI, and it shouldn't be.

One analogy landed particularly well:

"We've swapped the steam engine for a Formula One engine, but we're still relying on Victorian brakes."

The discussion wasn't about removing governance. Quite the opposite.

The consensus was that organisations need more governance, but of a different kind. Governance that is embedded into systems, platforms, workflows, and delivery processes rather than layered on top as manual oversight. 

The room largely agreed that the organisations that succeed over the next decade will be those that redesign trust itself rather than simply adding more checkpoints.

Interestingly, the question of accountability generated far less disagreement than expected.

While AI may generate code, accountability remains a human responsibility.

Several leaders made the point that organisations should not confuse capability with accountability. Human in loop should not translate to Humans holding all the keys to all the path to prod gates. Humans remain responsible for designing compliant systems, establishing appropriate controls, and defining governance models. The mechanics of generating, testing, deploying, and even recovering software can increasingly be automated, but responsibility for outcomes still sits with leadership.

The challenge isn't keeping humans in every operational loop. It's building the right loops, ones that reward good system behaviour and correct it when it falls short, instead of using human gates as a substitute for trusting the system in the first place.

In other words: break the agent-autonomy dilemma safely, instead of kicking the can down the road.

The challenge is redefining what meaningful human oversight looks like in an AI-native environment.

Measuring AI ROI Is Hard Because Measuring Value Has Always Been Hard

The second discussion focused on economics, but the conversation quickly moved beyond AI costs and model pricing.

A provocative question framed the discussion:

What happens first: AI delivers measurable business value, or organisations hit the limits of what they can afford to operate?

Interestingly, most leaders admitted they are not yet treating AI ROI as a prerequisite for adoption.

That doesn't mean costs don't matter. It means many organisations are still trying to understand the scale of the shift before optimising it.

One guest captured the mood perfectly:

"Asking organisations to prove the ROI of AI today feels a bit like asking Blockbuster to calculate the ROI of moving from VHS rentals to streaming. What happened to them? "

The room largely agreed that AI represents a broader strategic transition rather than a discrete technology investment.

The more interesting discussion centred on a difficult reality:

Most organisations still struggle to measure value creation across software delivery, regardless of whether AI is involved.

Many engineering organisations continue to measure outputs rather than outcomes. Delivery metrics, utilisation figures, and project reporting are abundant. Direct links between engineering activity and business value are often much harder to establish.

One guest observed:

"Attributing ROI to agents may eventually become as difficult as attributing ROI to individual employees."

Another reflected on the challenge facing finance teams:

"Most financial reconciliation systems are already complex, heavily manual, and often rely on data that has been massaged multiple times before it reaches leadership ROI dashboards and ‘business cases’.

The conversation moved toward a broader conclusion.

Before organisations can truly understand AI ROI, they need better visibility into value flow, profitability, and how work moves through the organisation. Future finance teams and CFO organisations may need to become far more context-aware, using AI themselves to understand where value is being created and how technology investments contribute to business outcomes.

The challenge isn't simply measuring AI.

It's measuring value.

The Future of Engineering Teams Is Being Oversimplified

The final discussion explored what separates organisations that successfully scale AI from those that don't.

Predictably, the conversation turned to talent.

Would organisations stop hiring junior engineers?

Would engineering teams become dramatically smaller?

Would AI eventually replace large parts of the software delivery workforce?

The workforce discussion surfaced one of the more interesting tensions of the evening.

Several guests pointed to a growing belief that AI will eventually lead organisations to hire only senior engineers. On the surface, the argument sounds logical. If AI can handle more of the execution work, perhaps organisations simply need experienced engineers directing the agents.

But the room quickly challenged that assumption.

"If we stop hiring junior engineers, where exactly do we think future senior engineers come from?"

The discussion highlighted a risk that isn't being talked about enough. Engineering talent isn't something organisations consume. It's something they cultivate.

Today's senior architects, engineering managers, and CTOs were all junior engineers once. Remove the apprenticeship model entirely and organisations may find themselves creating a talent shortage of their own making.

The consensus wasn't that junior roles disappear.

Rather, the nature of engineering careers is likely to change.

The more useful question isn't whether organisations will hire juniors or seniors. It's whether organisations are redesigning what those roles mean in an AI-native world.

Several guests argued that too much of the public conversation focuses on headcount reduction and not enough on capability evolution.

The future engineer may spend less time writing boilerplate code and more time designing systems, validating outcomes, orchestrating agents, understanding business context, and making judgement calls that AI cannot.

As one guest observed:

"We're not replacing the career ladder. We're rebuilding it while we're still climbing it."

The organisations that succeed will likely be the ones that intentionally develop the next generation of engineers rather than assuming AI will somehow do that work for them.

"The IT department of every company is going to be the HR department of AI agents in the future... Today, they manage and maintain a bunch of software from the IT industry. In the future, they'll maintain, nurture, onboard, and improve a whole bunch of digital agents and provision them to the companies to use."

— Jensen Huang, NVIDIA CEO

The Part That's Actually Hard

As happens at most EngineeringX dinners, the conversation eventually moved beyond technology.

That's when it became interesting.

Several guests spoke about the growing reliance on consultancies to help organisations navigate AI transformation. While most agreed external expertise can be valuable, there was a shared concern that some organisations are treating AI strategy as something that can be outsourced.

The room pushed back on that idea.

Technology implementation can be delegated.

Organisational evolution cannot.

One guest summed it up bluntly:

"You can hire someone to help build the roadmap. You can't hire someone to change your culture for you."

Another theme emerged around executive pressure and AI adoption.

Many leaders described receiving top-down mandates driven less by strategy and more by fear of missing out. The pressure to demonstrate progress is real. The challenge is that organisational readiness often isn't.

The phrase "culture eats strategy for breakfast" made an appearance more than once during the evening. Overused perhaps, but nobody disagreed with the sentiment.

The leaders seeing the most progress are not forcing AI adoption onto engineering teams.

They're creating environments where talented engineers can safely experiment, learn, challenge assumptions, and help shape the future operating model themselves.

One guest captured it particularly well:

"The people closest to the work need to become architects of the transition, not passengers on it."

The discussion also challenged the assumption that AI transformation is primarily a cost reduction exercise.

The leaders around the table repeatedly returned to the idea that governance, economics, talent, culture, and organisational design are interconnected problems.

Companies that optimise only for speed may create governance failures.

Companies that optimise only for cost may create capability gaps.

Companies that optimise only for compliance may slow themselves into irrelevance.

The organisations most likely to succeed are those willing to redesign the entire system rather than optimise individual parts.

One comment drew laughter around the table while capturing the seriousness of the moment:

"We're a bunch of plumbers trying to retrain as electricians. The problem is that nobody has fully thought through what happens when electricity meets water."

The laughter suggested the analogy hit close to home.

Designing the Future Before It Arrives

As the evening came to a close and conversations continued long after dessert, one thing felt increasingly clear.

The biggest risk facing organisations isn't that they'll move too slowly with AI.

It's that they'll move quickly without redesigning the systems, culture, and operating models needed to support it.

Throughout the evening, guests kept returning to a simple observation: every major technology shift promises transformation, but very few organisations are willing to transform themselves.

AI may be the most powerful toolkit engineering has ever received.

But toolkits don't create advantage.

People do. Culture does. Operating models do.

The organisations that thrive over the next decade won't be the ones that adopted AI first. They'll be the ones that had the courage to rethink how software is delivered, how trust is created, how value is measured, and how engineering teams are designed.

Because ultimately, this isn't a story about artificial intelligence.

It's a story about organisational intelligence.

And that is considerably harder to automate.