Sougata Roy
This work comes from the accountability layer beneath deployed AI. Who authorized an agent to act. What it was permitted to do. Who answers when it gets something wrong. The questions an examiner asks, written down before the examiner arrives.
The last twelve were spent inside regulated environments where governance is a legal requirement. For the past several years, the work has included AI agent deployment in federal regulated environments.
The frameworks here exist because those deployments required them.
Perspective
Experience
26+
Years building enterprise systems
Focus
Regulated AI governance
Currently deploying AI agents in federal regulated environments
The Deployment Origin
Governance became real when the agent went live.
Twenty-six years of enterprise systems work shapes this site. Twelve of those years were inside environments where governance was not optional.
The current work sits inside federal regulated environments where AI deployment decisions carry compliance weight. That level of detail is enough. The systems, agencies, and operational facts stay out of the public record.
The work revealed a recurring gap. Teams thought their governance covered the deployment because logs, access controls, and review processes existed. When the accountability question arrived, the written record was often thinner than the control stack.
The frameworks on this site come from that gap. They are not commentary from outside the room. They are the patterns that become visible when deployment, review, and accountability have to work together.
The Intent Architecture Stack came from environments where the accountability question had no clear answer and someone had to build one. It gives the authorization record a structure before an agent goes live.
The Accountability Assumption names a pattern seen repeatedly in regulated settings. Monitoring existed. Evidence existed. The missing record was the one saying who authorized the agent to act.
Forty minutes to write the record can prevent months of repair after the review.
Where The Work Lives
Free to read and cite. Built to be attributed.
The research lives in three places: The Governance Gap newsletter(opens in new tab), published every Tuesday on LinkedIn. Nine operational frameworks for regulated enterprises building the accountability layer for AI deployment. And an intelligence feed of dated, sourced observations on what is actually shipping in the Microsoft AI stack and what it means for governance. Everything here is free to read and cite. Use the frameworks and terms in your own work with attribution to Sougata Roy and sougataroy.com. Please do not republish, rebrand, or claim authorship of any framework, term, or model as your own.
The Governance Gap newsletter tracks these questions through primary sources only. It now includes more than thirteen editions written for technology leaders who have to govern AI after deployment.
The white paper, Who Owns the Agent?, publishes the Intent Architecture Stack as a formal accountability pattern. It is archived on Zenodo with DOI 10.5281/zenodo.20481551 under CC BY 4.0.
The site now contains nine frameworks and more than forty intelligence notes. Each one is part of the same operating question: what evidence survives when an AI deployment is examined?
Newsletter
Weekly writing
Ongoing observations published every Tuesday on LinkedIn.
Frameworks
Operational models
Nine operating models for accountability, authorization, and governance design.
Intelligence
Dated field notes
More than forty sourced observations on AI governance, security, compliance, and accountability.
Everything here is free to read and cite. Use the frameworks and terms in your own work with attribution to Sougata Roy and sougataroy.com. Please do not republish, rebrand, or claim authorship of any framework, term, or model as your own.
Contact
If something here landed for you, if a framework named a problem you have been carrying or an essay described a meeting you have already been in, I want to hear what it was.
Questions about the research or the governance questions it raises are welcome by direct message on LinkedIn.
Views expressed are my own and do not represent my employer.