Don Sanjeewa Alwis works across AI, digital transformation, research and technology entrepreneurship. His perspective connects the adoption of new capabilities with the readiness of the people who will use them.
Begin with a decision worth improving
Start by describing a process, the people it serves and the problem they experience. Establish what happens today before selecting a model or platform. A useful proposal explains both the potential benefit and the consequences of a wrong result.
Assess readiness across four areas
- Work: define the task, its boundaries and the decisions that remain with people.
- Information: check whether the necessary data is suitable, accessible and permitted for the intended use.
- People: identify the skills, review time and support the team needs.
- Accountability: name the process owner, escalation route and stop conditions.
Make the pilot measurable
For a hypothetical document-summary assistant, measure factual errors, reviewer effort and usefulness alongside speed. A faster draft is not a better process if verification takes longer or important details disappear. Compare the complete workflow with the starting baseline.
Evidence and related work
Sanjeewa’s co-authored research on Sinhala language model adaptation examines the practical constraints of developing AI for a low-resource language. His international engagements also include enterprise AI discussions in Finland.
Connect the pilot to its operating environment
Cloud infrastructure, identity, data architecture and operations shape what an AI system can safely do. Before connecting a pilot, document the information it needs, the permissions it receives and the route for revoking access. Separate read access from permission to change a record or trigger an action.
Consider two hypothetical checks. A document assistant should retrieve only information its user is entitled to see. A workflow agent that proposes an account change should require an accountable approval before execution. Both need records that let the team reconstruct an error and a practical way to stop the workflow.
These are design questions, not claims about a completed client implementation. Evaluate them alongside reliability, cost and the effort people need to review outputs.
Discuss your context
For a professional conversation, describe your organisation’s challenge, who is affected and what you need to understand before making a decision.