Invent & Simplify
I'm Koushal Dutt. I spent two decades delivering change in banking, at HSBC and First Abu Dhabi Bank, before moving to AWS in Dubai, where I help enterprises adopt cloud. I'm not a career developer: I now build AI systems with AI, and write about what works, what breaks, and the trade-offs in between.
More about me Views are my own, not my employer's.
Choosing an AI partner is a leadership decision. It influences which problems receive attention, which assumptions get challenged and how the organisation approaches change.
The studies claiming 10–30% productivity gains from AI coding tools are not wrong — they are just measuring the wrong thing. The real story is about what developers now attempt, not how fast they complete what they were already doing.
MCP's 2026-07-28 revision removes sessions from the protocol — but the state doesn't disappear, it moves through the client as an integrity-protected blob you now have to defend. What actually changed, what the coverage got wrong, and which kind of server you need to know you're building.
KAIROS is a personal AI operating system built on a Neo4j knowledge graph, MCP integrations, and agents on Amazon Bedrock AgentCore. Here's the stack, the design decisions, and what broke along the way.
Deploying AI tools shouldn’t require a DevOps degree. This guide shows developers how to deploy custom tools built with the Model Context Protocol (MCP) to Amazon Bedrock AgentCore Runtime in under 3 minutes. You’ll learn how to set up OAuth2 authentication with Amazon Cognito, launch a fully scalable HTTPS endpoint, and monitor performance—all without managing servers, containers, or scaling logic. Whether you’re building your first tool or scaling a fleet of agents, this is the fastest way to go from local script to production-ready AI infrastructure.
A grounded look at why AI feels so noisy, the seven dynamics behind hype fatigue, and what will last: products that are useful, durable and hard to copy.
Werner Vogels' six lessons on managing complexity from re:Invent 2024, from evolvability to cells and automation, plus how Canva and Too Good To Go scaled.
Essential best practices for building production-ready infrastructure with AWS CDK.
Why decisions go wrong through ignorance or arrogance, the seven traits behind bad judgement, and the values, reflection and empathy that lead to sound ones.
How to persevere at work: build a vision from your own passion, stay focused on details, work hard without burning out, and adopt seven practical habits.
Why cloud is central to digital transformation: scalability, low-cost experimentation and data-driven decisions, with retail, startup and healthcare examples.
A step-by-step way to manage security, compliance, operational and financial risk in banking cloud projects, with AWS services for each stage.
How the AWS Cloud Adoption Framework and Well-Architected Framework work together: one readies your organization, the other keeps workloads to best practice.
Personal project, 2026
A personal AI system: a Neo4j knowledge graph fed through MCP integrations, with Strands agents running on Amazon Bedrock AgentCore.
Keeps working context (meetings, people, commitments) across days instead of starting every conversation from zero.
This site
A statically exported Next.js site on S3 and CloudFront, with Lambda and DynamoDB APIs, all defined in AWS CDK.
Built and run with AI as the main collaborator, by someone who is not a career developer.
Head of Enterprise and Group Projects, 2021–2022
Led cloud modernization in a $150M technology portfolio, including an IBM to AWS migration.
Cloud adoption inside a regulated bank, where governance and risk decide the pace as much as technology.