There’s a familiar conversation in data consulting. A customer has a budget, a team and a long list of things the business wants from its data. More reports. More trusted data assets. Better access to information that helps people make decisions.
Then we work through what’s achievable with the people, time and money available, and that list gets smaller.
The gap between what organisations want from their data and what they have the capacity to deliver has always shaped our work. AI gives us a way to close more of that gap.
A customer might still come to us with the same budget. What’s changing is how much value we can deliver within it. Work that previously required a large team and a long delivery window becomes achievable with fewer people, using agents to take on more of the development effort.
That’s a significant opportunity for data teams. It also raises a question I think we need to spend more time on: what exactly are we handing over to the customer?
The ability to build more must come with an understanding of what they can operate afterwards.
Agree on what done means
As agents take on more development responsibility, we’re going to see different expectations of the finished product.
Some customers will want AI to help us deliver in a way their teams already understand. Their people will review the code, work within familiar engineering patterns and take responsibility for maintaining the solution. AI accelerates the work, but the output still needs to be built for human consumption.
Other customers will be ready for a delivery model where agents do more of the building and ongoing maintenance. They’ll expect an output that has been tested, validated and proven to meet agreed requirements, without necessarily expecting a human to review every individual change.
Those expectations need to be understood before delivery starts.
If we turn the tap on and generate a large amount of code, then discover the customer expects their team to review and maintain every part of it manually, we’ve created a problem. Equally, if a customer is equipped to work agentically, putting a manual review step around every change may prevent them from getting the speed they came for.
The conversation needs to happen early. What is the definition of done? How will we demonstrate that the solution works? Who will maintain it, and how will they make changes?
The answers should shape the delivery approach.
Stepping back allows you to see more
When you’re writing code yourself, you’re close to every detail. When someone else does the development and you review their work, you take a step back.
AI allows that process to continue. As you build the skills, frameworks and patterns that consistently produce a quality output, you can step back further. That gives you more room to look across the solution and understand how the pieces fit together.
There are still people involved. The question for operational leads and platform owners is where their attention has the most value, and how much detail needs human review.
An agentic approach still needs standards. You need to test the output, execute it, validate its behaviour and establish that it meets the characteristics you’ve agreed with the customer. Moving a person out of a particular development step means having a dependable way to establish quality at that step.
How far you can take that depends partly on the customer’s ability to keep working that way once the engagement ends.
Build for the team taking over
AI makes it possible to build some extraordinarily capable and complex solutions. As development capacity increases, the limits become more about attention, imagination and budget.
But a customer’s capacity to support the result doesn’t automatically increase with it.
If we deliver a larger, agentic codebase with more functionality, then hand it to a team that operates manually, we may have given them more than they can reasonably maintain. The project might meet its delivery requirements while setting the existing team up for failure.
Sometimes getting the balance right means exercising restraint. In consultation with the customer, we need to make conscious choices about how much complexity a particular solution should contain, at that stage of theirjourney.
Sometimes it means helping the organisation change how it builds and operates, so its own people can use agents effectively too.
At Arinco, that’s an important part of how I think about agentic data delivery. We need to understand the customer’s operating capability alongside the immediate project requirements. Where they want to move towards a more agentic model, helping them develop that capability is part of making the solution useful over time.
The value has to continue after we leave.
Data platforms are becoming foundations for agents
There’s another side to this change. AI is helping us build data platforms faster, and those platforms are becoming the foundation for what organisations want to do with AI.
Businesses want to understand their data assets, build applications on top of them and use that information to support operations and decisions. That might include proactive monitoring, alerting or agents that help carry out business processes.
High-quality, well-governed data assets make those possibilities much more useful. They give the organisation a dependable foundation to build on.
They also make the controls around that data more important.
Despite clear guidance, risks, controls and recommendations, most organisations continue to operate with users with over-privileged access, , trusting the people operating their systems to not do dumb things. They trust the person with access to production because they understand the environment and the consequences of making a mistake.
That assumption absolutely does not carry across to an agent.
Assume that if the agent has access, then eventually it will do the thing you didn’t intend. Solution access, testing and operational controls need to keep the agent restricted and sandboxed, so the risk is mitigated.
Can it access the wrong information? Could it share that information with the wrong people? What actions can it take, and what happens when something goes wrong?
The requirements are familiar. Solutions still need to be designed, developed and tested so they’re safe, secure, efficient and reliable in production. The ability of agents to carry out more work makes applying those principles consistently even more necessary.
Make the extra capacity count
For customers, for the first time ever, there is the capacity to meet the businesses data needs. Until now, pretty much every organisation has had its data ambitions constrained due to lack of people or budget.
More insights the business needs. More trusted data products. More modernisation work that becomes achievable within an available budget.
My view is that we should pursue that opportunity with a clear understanding of what happens after delivery. Agree on the output, establish how quality will be demonstrated and build for the people who will run it.
If you’re looking at a data backlog that has always felt bigger than your team or budget, it’s worth revisiting what’s possible. At Arinco, that conversation starts with the outcomes you need and how your team will operate the result. From there, we can work out how agentic delivery can help you achieve more.


