The barrier to building software has essentially collapsed.
That's not hyperbole. A junior analyst with a clear prompt can now have a functional data pipeline running in an afternoon, a task that would have taken a senior engineer most of a week. For business leaders, it genuinely feels like you've cracked the ultimate productivity code. For data leaders, that same excitement should come with a degree of caution.
The more AI lowers the cost of building, the more gets built
In the 19th century, economist William Stanley Jevons observed that when steam engines became more fuel-efficient, coal consumption didn't fall; it skyrocketed. When something becomes cheaper to produce, people don't produce the same amount more efficiently; they produce vastly more of it.
Software follows the same logic. As AI removes the barriers to writing code, every team suddenly wants bespoke tooling, every workflow becomes a candidate for automation, and every department has an idea for an app. The volume of software expands faster than any organisation's capacity to govern it, and that gap is precisely where the risk lives.
"AI slop" at enterprise scale
When this happens, you risk generating what is commonly referred to as 'AI slop' on a massive scale. To be clear: this isn't a comment on whether the code works, it often works very well. The problem is that working code and governable code are two very different things.
AI slop is what you get when you have transformations that nobody can trace back to a business rule, pipelines that nobody can explain to an auditor, models that nobody can validate, and architectures that satisfy a prompt but break a regulation. If your teams are using AI to build at any kind of scale right now, some of this is almost certainly already in your estate.
The reason is structural, not careless. AI generates boilerplate syntax brilliantly, but what it inherently lacks is what engineers call mechanical sympathy: an awareness of the infrastructure it's operating on, an understanding of downstream dependencies, and any concept of what a regulator might ask about it in six months' time. Every pipeline you can't explain to an auditor is a risk you're carrying on your balance sheet. Speed without governance isn't a feature; it's a liability that compounds quietly until it isn't quiet anymore.
From code writers to system orchestrators
The organisations that extract real, durable value from AI aren't the ones that build fastest; they're the ones that think before they build. Software engineering is no longer just about writing code. It is a discipline of managing complexity. The constraint has moved away from syntax and towards judgement, governance, and architecture, which are fundamentally human capabilities.
This reshapes the data engineer's role significantly. AI now absorbs a large portion of what used to define that job: writing boilerplate, building routine pipelines, handling the mechanical parts of development. What remains is considerably more valuable: complex orchestration, resilient architecture, enterprise governance, high-level system design.
This view is gaining traction across the field. Researchers at Google DeepMind have made the same observation: that while enormous effort has gone into improving the underlying models, the critical next step is learning to coordinate, orchestrate and manage them effectively, and that people need to start seeing themselves as managers of teams and institutions rather than authors of individual solutions. The engineers who thrive in this environment are the ones who can direct AI effectively, design systems that govern it, and understand the difference between code that works and code the business can actually own. We describe this shift as moving from code writers to system orchestrators, and for most data teams it isn't an optional evolution; it is simply what the job now requires.
Why individual thinking isn't enough
There is, however, a trap in taking that message back to your organisation and leaving each team to work out their own approach. If every function develops its own quality checks, prompt patterns, and security model in isolation, you end up spending every hour AI saved you just recreating the same guardrails from scratch, over and over again. As AI agents take on more complex knowledge work, siloed governance doesn't protect you; it undermines the efficiency you were trying to achieve in the first place.
The real multiplier comes from codifying good thinking into shared infrastructure: reusable agents, proven prompt patterns, standardised quality gates, and workflows that are both repeatable and auditable. The new competitive asset isn't raw data, and it isn't AI output either. It's structured context, the codified judgement of your best people, embedded into the platform so that every team can benefit from it, not just the ones lucky enough to sit alongside your most experienced engineers. That is what converts a one-off AI success into a genuine organisational capability.
Engineering the central nervous system
This is where the mandate of the data platform team has to fundamentally change. Provisioning data was last decade's job. The mandate now is to engineer the central nervous system for AI across the enterprise, which means taking ownership of four things:
- Secure, governed pipelines: the data your agents operate on is clean, consistent and controlled before it's used
- Identity and access controls: you know precisely which agents are doing what, with whose authority.
- Guardrails: AI cannot execute commands your business hasn't explicity sanctioned.
- Observability: when something goes wrong, you can see it, trace it, and explain it to anyone who asks.
There is also a behavioural dimension to governance that technical controls alone cannot address. Google DeepMind researchers describe this as automation bias: when agents perform reliably over time, the people overseeing them gradually disengage, and subtle errors that would previously have been caught start to slip through. The four capabilities above create the structural conditions for safe AI operation; keeping people genuinely engaged in oversight is what makes those conditions effective in practice.
None of this is optional. Without these foundations, what looks like an AI capability quickly becomes an AI liability. Every autonomous agent that touches enterprise data should be required to pass through all four layers before it executes anything, and this is an architectural decision that has to be made at the platform level rather than left to individual teams to resolve for themselves.
What data leaders should be focused on right now
When this foundation is in place, AI investment delivers sustainable returns. Speed and cost efficiency matter, but the greater value lies in institutionalising expertise. The hardest-won judgements of your best engineers—how to safely prompt an agent, how to build auditable pipelines, and how to design for failure—become reusable assets that can be shared across the organisation and retained even as underlying models continue to evolve.
Cost is becoming an equally urgent dimension of this challenge. AI inference is consumption-based, and the same Jevons dynamic that drives software proliferation drives spend proliferation. As individual teams spin up their own agents and model integrations, AI costs fragment across the organisation in ways that are difficult to attribute and harder still to justify at board level. The organisations that build centralised, observable AI infrastructure don't just govern quality and security more effectively; they also gain the visibility needed to understand what AI is actually costing, identify which use cases are generating genuine value, and answer with confidence when the CFO asks what the AI budget is delivering. As boards begin asking sharper questions about return on investment, that visibility is shifting from a nice-to-have to a strategic necessity.
The central challenge for data leaders right now is converting AI from a black-box liability into a governed, repeatable asset the business can own and explain. The organisations that build this foundation will move faster, with less risk, and with something competitors cannot easily replicate. In a world where anyone can build, the real advantage belongs to those who build in a way that can be understood, trusted, and sustained over time.
At Snap Analytics, we help organisations accelerate data delivery without losing control, embedding the governance, architecture and reusable foundations needed to scale safely and sustainably.
If you’re looking to move faster with AI-enabled data delivery, book a discovery call with our team to explore what that could look like for your organisation.