Fri, 20 Nov 2026

The AI-ready enterprise: What to consider before scaling AI

https://pixabay.com/illustrations/office-meeting-business-7853103/

Successful AI scaling requires more than investing in new models and applications. Organisations must first address infrastructure, data, and security challenges that impact AI’s business value.

“I think of AI as an amplifier,” said Adam Seyer, chief information officer at StarHub.

Adam Seyer

AI can enhance an organisation’s capabilities but may also reveal weaknesses in its technology foundations. Strong engineering, robust cybersecurity, and a disciplined software development lifecycle make AI more effective, while poor documentation, weak quality controls, and unclear processes can undermine initiatives.

“The challenge organisations need to overcome first isn’t really about AI. It’s the fundamentals, and the effort you put into them compounds,” Seyer said.

The challenge organisations need to overcome first isn’t really about AI. It’s the fundamentals, and the effort you put into them compounds. Adam Seyer

Focus on the fundamentals

For organisations aiming to scale AI, especially those not built as cloud-native, Seyer highlights two key fundamentals: ownership and standardisation.

The first is taking ownership of the systems they build and operate.


Many established businesses rely on multiple external vendors for technology implementation and maintenance. Over time, this can place critical operational knowledge outside the organisation, making it harder for internal teams to understand, modify, or improve systems.

“When that’s the case, AI has nowhere to land: you can’t redesign a process you don’t understand, on a system you don’t control, through a contract that was written before any of this existed,” he said.

The second is establishing what Seyer calls a “paved road”: a limited set of agreed-upon, supported methods for building and operating services, agents, and applications.

This includes hosting environments, deployment pipelines, monitoring, and security controls. Instead of making individual infrastructure decisions, engineers can focus on solving business and customer problems.

This is even more important with AI, which enables development teams to produce much more code. Without standardised infrastructure, each new application could introduce additional systems requiring security, monitoring, and maintenance.

StarHub is automating this process, aiming to enable engineers to move from concept to a running service with required infrastructure and controls in under an hour.

Data foundations are equally important. AI systems may generate confident answers from inaccurate, untraceable, or inconsistent data fields. Understanding data origin, ownership, and meaning is essential before making it available to AI models.

Security must underpin these foundations from the start. As a critical information infrastructure operator, StarHub cannot treat governance as an afterthought. Establishing controls early is essential for using advanced models in sensitive environments.

Making data usable

Another challenge for established organisations is making years of business data accessible to AI without overhauling existing IT infrastructure.

Seyer notes that organisations do not always need to replace legacy systems to achieve this.

“Some of ours we’ll never rewrite, and that’s a perfectly sensible decision. What we change is the layer around them: how data gets out, who can reach it, and through what interface.”

AI can also help organisations understand and work with older systems.

Legacy applications often contain years of code that may be poorly documented or understood by few employees. AI can analyse this code, explain system functions, and support development of cleaner interfaces for integration.

Instead of consolidating all enterprise data into a single platform first, Seyer recommends a more incremental approach.

“Pick a real problem, expose the data that problem needs through a governed interface (standards like MCP (Model Context Protocol) and A2A (Agent2Agent) are making this much easier), and let the next problem reuse it.”

This approach helps organisations build a network of secure, well-understood connections to legacy systems. Each integration becomes a foundation for future AI initiatives, reducing effort for subsequent projects.

Data governance and access controls

AI doesn’t need a new security model. It shows you how good your existing one is. Adam Seyer

Making enterprise data accessible to AI also requires ensuring AI applications access and process only authorised information.

Seyer believes this does not require a new security approach. Instead, AI reveals the strengths or weaknesses of existing security practices.

“AI doesn’t need a new security model. It shows you how good your existing one is,” he said.

AI can highlight previously overlooked access issues by surfacing overly shared information or identifying documents employees may not have realised were accessible.

“Before worrying about the model, fix who can see what. If your identity and access controls are sloppy, AI will find every gap,” he said.

Seyer shared that their company follows a principle: treat an AI agent like any other identity in the organisation. It should never access more information than the person it represents.

This means avoiding overly privileged service accounts for AI deployment. Every model call should pass through a governed layer that provides visibility into accessed data, involved tools, interaction purposes, and associated costs.

Clear accountability is also essential. At StarHub, data protection, legal, and security leaders oversee policy, with the CISO shaping the approach.

Technology teams embed these policies into the platform, rather than relying solely on written guidelines.

For CII operators, the stakes are high. Avoiding advanced models in sensitive environments may seem safer, but Seyer believes this is unsustainable and could leave critical systems dependent on less capable tools.

Moving beyond AI-enabled workflows

The right infrastructure and governance are only part of the solution.

Organisations must also assess whether existing applications and workflows need to change to realise AI’s full potential.

Seyer believes businesses should look beyond simply adding AI to current processes.

“Most organisations add AI to an existing workflow. The bigger gains come from asking whether the workflow should exist at all,” Seyer shared.

Many established business processes, such as manual data entry, reconciling figures across systems, and coordinating handovers, exist because older technology could not automate them effectively.

Adding AI to these processes without re-evaluating them may only accelerate inefficiency.

Software development is one area where StarHub has made significant changes.

In redeveloping a consumer application, the company uses AI for the full development lifecycle, while engineers define requirements, approve work, and remain accountable for releases. Code is cheaper and faster to produce; the bigger challenge becomes deciding what to build and ensuring it works as intended.

This makes testing and quality assurance even more important.

Organisations that underinvest in automated testing may find AI-generated code exposes weaknesses in their development practices.

Seyer emphasises that supporting tools and processes are as important as the AI model itself.

A model may generate useful answers, but infrastructure, permissions, and workflows determine whether those answers become secure, operational business outcomes.

Ready to scale?

As AI tools become more accessible, adoption alone is not a reliable indicator of enterprise readiness.

Seyer believes organisations should instead assess how effectively they move new ideas into production.

Seyer noted: “Counting AI tools tells you almost nothing about readiness; it mostly tells you about procurement. If I could only look at one thing, it would be how long it takes to get a new idea into production safely.”

The time needed to deploy a new application or service reflects underlying capabilities, such as infrastructure maturity, testing effectiveness, and whether security is embedded from the start.

If deployment takes months, AI may only help teams produce more work that remains stuck in the pipeline.

Beyond deployment speed, leaders should evaluate how AI changes organisational behaviour. This includes tracking the cost and usage of AI initiatives, discontinuing pilots that lack value, and determining whether AI is transforming core operations or just appearing in demonstrations.

Flexibility is also important. Organisations should be able to switch between AI models without rebuilding their entire technology environment.


As models evolve, relying too much on a single provider could limit an organisation’s adaptability.

Scaling AI successfully

Ultimately, scaling AI is less about adopting the latest technology and more about building an enterprise that can use it effectively.

Organisations that invest in strong engineering, accessible and governed data, security, and adaptable infrastructure will be better positioned to turn AI experimentation into sustainable business value.

Related:  Business technologists drive enterprise-wide tech investments

Related Stories

MORE STORIES

Subscribe