Yasir Qureshi
I run large programs. I also write when I'm trying to understand something.
I grew up in Karachi and now live in the Toronto area. Most days I work in the Corporate PMO at Empire Life.
Outside work, I read and write. Sometimes a question bothers me enough that I design a system around it: the controls, specifications and acceptance criteria. I use AI to implement the prototype and verify the resulting behaviour.
I am a father and a Barcelona supporter. I still think in Urdu first.
Right now
I am implementing Workday within Empire Life.
Selected work
Three pieces of work, with the part I was responsible for.
Current
Corporate PMO · Empire Life
I am Empire Life's internal project lead for Projects, Time and Expense, and Procurement. KPMG leads the wider implementation; I manage Empire Life's side of these workstreams and am helping Corporate Finance through its first quarter-end on Workday. I also managed KPMG's work in Workiva to automate most fund reporting and add cross-checks.
~$2M / year
Product · Actuarial · Group leadership
Our calculation used a percentage where Quebec's drug-insurance pool used a dollar factor for each certificate. I worked with Product and Actuarial to change the method. We estimated the correction would stop roughly $2 million a year in overpayments.
Life + disability
Enterprise claims system
I led the selection and implementation of Majesco's Salesforce-based claims system, replacing a partly manual legacy system. I ran the RFP and RFQ, vendor negotiations, and delivery with the business and technical teams.
Writing
Seven essays about the trust problem around AI agents. I do not have a complete answer. The essays are me trying to work one out.
“The defender has the same architecture as the threat.”
I wanted a browser extension to warn people about bad terms of service. Then I realized it needed the same kind of access I was warning them about.
Start with The Trust ParadoxLab
Small systems I made because I needed to see whether an idea would actually work. I design the controls, specifications and acceptance criteria, use AI to implement the prototypes, and verify the resulting behaviour.
A month-end close experiment where the AI could read and advise but could not change a number.
ArchivedA small console built around one idea: an AI agent should order from a menu instead of carrying the keys.
Working demoAn attempt to give an AI agent a memory that says “I don’t know” instead of filling in the gap.
In progress