TechBOT Talks
1. Enterprise AI adoption has moved quickly over the past few years. From what you are seeing on the ground, what is changing in the way enterprises now think about AI?
The shift I see in boardrooms is from \what can AI do?\ to \what can we trust it to do in production?\ Many enterprises have run pilots and rolled out copilots. Far fewer have taken AI into the core of how the business runs. The questions have changed. What can an agent do on its own? Where must a person step in? Who is accountable? Can every outcome be explained? Copilots made individuals faster. Enterprises now want whole processes to run better, and that depends on governance and outcomes more than on the model.
2. What is the biggest challenge created by having multiple AI tools and models across an organisation?
I call it AI sprawl. A voice bot in one function, a coding assistant in another, teams using consumer tools on their own. Each solves a local problem. Together they create silos, scattered data and no single view of who approved what. There is also a growing volume of AI-generated code that nobody planned for and few can fully account for. Banning these tools is not the answer, because people will find a way. Give every use case a common, governed foundation and sprawl becomes a portfolio.
3. How important is integration with existing enterprise systems for scaling AI?
It is critical. AI that sits outside core systems stays a productivity add-on, however impressive the demo. The value shows up when it works inside the workflows that move money and carry risk, such as reconciliation, lending and compliance. In my conversations with banks the test is simple: does it change time to market and the cost of technology, or does it just make someone's morning easier? Integration is what moves those two numbers.
4. Why do legacy systems remain a major barrier to enterprise AI adoption, especially in BFSI?
Most financial institutions run lending, deposits and payments on separately procured platforms, each with its own history and change process. They were never designed to change together or share signals in real time, so modernisation programmes stall for years. The answer is not rip and replace. Keep the core running and add a layer above it, where new capabilities are configured, approved and governed. Institutions can then launch in days what used to take quarters, without touching the core.
5. What should enterprises build underneath their AI tools to make deployments more connected and scalable?
Three things. A shared layer for identity, audit and version control. Common domain intelligence, so every use case draws on the same source of truth. And a clear separation between AI and production. AI should design and configure. A person approves, and deterministic engines execute. That is how we build MiFiX.ai. AI never runs uncontrolled in live operations, so every outcome is repeatable and auditable.
6. How can organisations ensure different AI systems work with the same data and business processes?
By not letting each tool build its own version of the truth. Give them a common, permissioned source of institutional knowledge, so a fraud model in retail banking and a compliance workflow in trade finance work from the same intelligence. Sensitive customer data stays inside the institution's governed environment. And configure processes instead of coding each one separately, so a rule changed once applies everywhere it should.
7. What are the key considerations around governance and security as AI moves deeper into enterprise workflows?
Governance is not the brake. It is what lets you go faster, because a board will only approve scale it can defend. Build it in from day one: audit trails, version control, maker-checker and human oversight. Let AI perceive, recommend and design. Let deterministic engines decide and execute, within rules and approvals. Assume the model will sometimes be wrong, and keep a kill switch. Above all, accountability stays with the board and leadership. Adopting AI does not transfer it.
8. What will enterprise AI architecture look like over the next two to three years?
I expect the pilot era to end. The measure that will matter is production-grade adoption: how much AI actually runs in live, governed operations, not how many pilots exist. Enterprises will connect their systems through shared, permissioned intelligence, and agents will begin to work with each other. In working-capital lending, one agent could gather data from a customer's ERP while another checks it against the bank's credit rules. The winners will be those who keep control, with explainable, auditable systems, clear accountability and a human in the loop where it matters.
