Kinetic IT urges AI continuity plans for critical sectors

Kinetic IT has urged organisations to build continuity plans into their artificial intelligence strategies, particularly in government, defence and critical infrastructure.

Kishore Jayaram, Chief Transformation Officer at Kinetic IT, said AI planning is moving beyond how quickly organisations can adopt new systems. Leaders must now consider whether AI-enabled operations can continue if the underlying model, provider or operating environment changes.

"Intelligence continuity is the ability to preserve and evolve important organisational capability as AI models, providers, and operating conditions change. While the AI model matters, it's just one component inside a much larger architecture. Organisations need to be able to preserve the outcome when the intelligence underneath it changes, without disrupting the critical services and capabilities that depend on it," said Kishore Jayaram, Chief Transformation Officer, Kinetic IT.

His comments reflect a broader debate about how much lasting advantage comes from access to large AI models as similar tools become available from multiple suppliers. In that environment, Jayaram said the more important asset may be the systems, workflows and organisational rules built around the model, rather than the model itself.

That surrounding environment includes internal knowledge, customer and operational context, workflow design, software integrations, permissions, controls, evaluation methods, decision records, human approvals and fallback arrangements. As organisations rely more heavily on AI, these layers can capture how work is done and how decisions are governed.

"It's a simple principle: borrow the intelligence, own the capability. Traditional concerns about vendor, contractual, jurisdictional, data, and model dependencies remain relevant. However, the harder dependency may emerge in the surrounding environment, where institutional knowledge and operating decisions become embedded in agents, workflows, tools, and controls. Intelligence continuity addresses the distinction between being able to switch to another model and being able to quickly restore the same operational capability," Jayaram said.

Control points

Most organisations will use AI models developed by others rather than build their own frontier systems. That leaves them with a different point of control: the framework that determines what an AI agent can access, which tools it can use, what actions it can take, and how those actions are recorded and reviewed.

Jayaram described that framework as a strategic asset because it embeds organisational judgement. One business may choose a tightly integrated single-provider approach for speed, while another may preserve flexibility across multiple providers to reduce concentration risk, even if that adds cost and complexity.

"There is no universal answer. The appropriate balance will differ between a commercial organisation and an agency responsible for public services, national security, or critical infrastructure. The important point is that the trade-off is deliberate rather than accidental," Jayaram said.

The argument becomes more pressing as AI agents move from assisting workers to taking actions inside business and operational systems. Once agents gain access to data, software tools and delegated authority, they function less like passive applications and more like active participants in a workflow.

Jayaram said this shift requires organisations to define identity, purpose, permissions, human sponsorship, delegated authority, performance standards, monitoring and decommissioning processes. The more context and authority an agent receives, the more useful it may become, but the consequences of failure can also rise.

"We can delegate actions to AI; however, we cannot delegate accountability. As agents take on more work, organisations need to engineer both the limits of what they can do and the points where meaningful human judgement is required," Jayaram said.

Four actions

Jayaram set out four measures for technology and operational leaders. The first is a "24-hour test", which asks what would happen if an important AI-enabled capability disappeared for a day. The issue, he said, is not how advanced the technology is, but what the absence of that function would mean for the organisation.

The second step is to create what he called a capability bill of materials. This involves mapping every dependency needed to sustain an AI-enabled outcome, including models, platforms, data sources, software tools, integrations, permissions, workflows and controls, to reveal where practical lock-in exists.

The third action is to decide which elements of the capability must remain under the organisation's control so the outcome can be rebuilt elsewhere if necessary. That may include business rules, institutional knowledge, workflow definitions, evaluation sets, context, controls, decision history and the evidence needed to explain a result.

The fourth measure is to set an intelligence recovery objective. In practice, that means deciding how quickly an acceptable substitute must be in place if a model or provider becomes unavailable, and whether manual or degraded modes are needed for critical functions.

Truly critical capabilities should have alternatives tested in realistic conditions, including changing the model, restoring workflows, reapplying permissions and rerunning evaluations to see what fails, he said.

"Agentic AI changes more than the technology stack, and as AI moves from assistance to execution, it can change the boundary between people and technology. These four actions turn intelligence continuity into an operating discipline for organisations to understand the consequence, know the dependency, retain the recipe, and invest in the resilience the consequence warrants," Jayaram said.

He added that organisations do not need certainty about which AI model or provider will lead the market in the near term, but they do need to understand which decisions made now will be hard to reverse later.

"Organisations don't need to predict which AI model or provider will dominate in two years; however, they do need to understand which decisions made today will be difficult to undo tomorrow. The organisations that get material value from AI will be those that turn borrowed intelligence into enduring organisational capability," Jayaram said.