Cubesys: Redesigning work for the age of AI

Q& A Avron Welgemoed, Strategic AI Advisor and AI Practice Lead at Cubesys

1. Where are organisations struggling most when it comes to scaling AI, and how are you helping address those gaps?

The biggest gap we see isn’t technology — it’s thinking. Most organisations are still treating AI as a smarter search engine, asking it questions one at a time, rather than understanding the flow of work through their business and where AI can act as a partner inside that flow. They’ll pilot a chatbot here, a copilot there, and then wonder why productivity isn’t actually moving.

That’s exactly why we built AI Forge. AI Forge is cubesys’ systems-thinking approach to AI adoption — we map the end-to-end process of how value moves through an organisation, then design where AI agents, Copilots, and automation belong inside that flow. The aim isn’t to bolt AI onto existing tasks; it’s to redesign the work so people and AI are doing what each does best. The gap closes when leaders stop asking “what tool should we buy?” and start asking “what is the shape of work in our business, and where does AI change it?”

2. Many organisations have moved to the cloud but still struggle to operationalise AI. What’s missing, and how do you help move from pilots to production?

Three things are usually missing: a clear line of sight from AI to a business outcome, the data foundations to make the AI trustworthy, and a People and Adoption plan that changes how staff actually work. Pilots stall because they’re treated as IT experiments rather than operating-model changes.

Through AI Forge we run a People and Adoption Act alongside every deployment — because the hardest part of moving from pilot to production is shifting mindset. People need to stop using AI like Google and start using it as a partner that does parts of their day-to-day job. That’s a behavioural change, not a training course. We pair that with the technical foundations — Microsoft 365 Copilot, Azure AI, and now Microsoft Agent 365 (launched 1 May) — to give us a managed control plane for the agents themselves. Agent 365 is a genuine game-changer because it gives us a way to govern, secure and observe AI agents the same way we govern users today, which is what production-grade AI has been waiting for.

3. What changes to architecture and data strategy are essential to support scalable AI?

Architecture has to follow the business process, not the org chart. If you can’t see how work actually flows through your business — across systems, teams, and data — it’s almost impossible to apply AI in a meaningful way. So the first step is mapping that flow.

From there, the essentials are well-aligned with Microsoft best practice: a unified data estate built on Microsoft Fabric and OneLake so AI is grounded in clean, governed business data; Microsoft Purview as the backbone for sensitivity labelling, data lifecycle management, DLP and information protection so AI can only see what it’s allowed to see; and an identity and endpoint posture uplifted so the controls scale with the AI surface area. Without the labelling and Purview work in place, every Copilot deployment surfaces oversharing risk on day one — that’s what derails most rollouts.

4. How are you helping organisations manage risk and build secure, responsible AI in cloud environments? 

We think about risk in four layers, and you have to get all four right — getting three out of four is what creates the headlines.

• The data layer — sensitivity labels, auto-labelling, DLP and lifecycle management through Microsoft Purview. If your data isn’t classified before Copilot or an agent reaches it, oversharing is inevitable. Purview is non-negotiable as the foundation.
• The identity layer — modern identity, conditional access, MFA, Defender for Identity and Defender for Cloud. The AI surface area only scales safely if your underlying identity and endpoint posture has scaled with it.
• The agent layer — every AI agent is effectively a new identity acting on someone’s behalf, so it needs to be registered, governed, observed and lifecycled. Microsoft Agent 365 gives us that control plane, and frankly it’s the piece the industry has been missing. It’s how we get from “agents in the wild” to agents you can actually audit.
• The human layer — the People and Adoption inside AI Forge. The most common security incident with AI isn’t a model breakout, it’s a person treating AI like Google and putting the wrong data into the wrong prompt. That’s a mindset problem, not a tooling problem, and the only way to solve it is to deliberately re-shape how people work with AI as a partner — what to delegate, what to verify, what stays human.

Responsible AI shows up when those four layers are designed together, not bolted on afterwards. That’s exactly the work we do with customers — and what Microsoft has built the platform to support.

5. Can you share a customer example that demonstrates impact, and what advice would you give CIOs looking to scale AI?

Let’s start with ourselves – cubesys is Customer Zero for AI Forge, and we eat our own dog food. We’ve put ourselves through the same end-to-end process we take customers on: mapping how work flows across the business, identifying where AI agents, Copilots, and automation actually fit, applying Purview and sensitivity labels to strengthen the foundations, and driving the People & Adoption piece so our team treats AI as a partner – not just a search engine.

That discipline is deliberate. It means every recommendation we give a CIO has been pressure-tested in our own environment first – including the awkward parts.

My advice to CIOs would be three things:

• Map the flow before you buy the tool. Understand how work actually moves through your business. AI value comes from redesigning the flow, not from sprinkling features on top of it.
• Treat People and Adoption as the project, not the wrap-around. The mindset shift from “AI as search engine” to “AI as working partner” is where the productivity gains actually live.
• Get your governance house in order — Lock in Purview, sensitivity labels, identity and endpoint security — before you scale. Then bring agents into a managed control plane like Agent 365. Don’t let your AI surface area outrun your controls.

Do that, and the shift from cloud-enabled to AI-enabled stops feeling like a leap – and starts becoming a system.