Anshul Kaushik

Managing Director

Anshul is the Managing Director at Ryzolv. He writes about AI strategy, governance, AI harnesses, managed automation and SAP.

His articles focus on the business process, the controls around AI and the decisions people remain responsible for.

Articles by Anshul Kaushik

Decide which agent actions need human-in-the-loop AI, who should approve them and how to enforce approval through identity controls.

Understand model-directed workflows, tool access, the surrounding AI harness and the limits that matter before deployment.

How to make approval a specific, inspectable decision rather than a generic checkbox after a model response.

Questions to ask about scope, data access, evaluation, professional review and operating responsibility before selecting an AI partner.

How to turn an AI initiative into a bounded project with requirements, data decisions, acceptance evidence and operating ownership.

Understand the organisation’s role, use-case classification, current application dates and the evidence an engineering team needs to prepare.

A clear way to compare AI tools by their access, actions and human responsibilities rather than by product labels.

How to maintain a common engineering control set while assessing the laws and operating requirements that differ by organisation and use case.

How inconsistent records, missing definitions and stale sources affect AI systems, and how to make those failures visible.

Private hosting changes the risk boundary. It still needs access control, egress review, artifact integrity and application-side safeguards.

Map personal information across prompts, retrieval, tools and logs so privacy obligations are assessed against the real AI workflow.

How fixed workflows and AI interpretation can work together without making every business action depend on a model.

How tool access, untrusted content and uncertain external outcomes change the risk of model-directed workflows.

How to use AI during SAP modernisation while preserving clean-core decisions, supported interfaces and release evidence.

Use dependency, usage and test evidence to guide AI-assisted analysis of custom ABAP without treating generated code as verified business logic.

How unclear ownership, weak acceptance criteria and missing integration work can prevent an AI pilot from becoming useful operational software.