Michael Anyfantakis, Co-Founder
I was there for the first wave of Digital Transformation in banking.
I’ve seen the 3 year roadmaps and 12 month plans, the exhaustive requirements gathering, and the extensive testing cycles. At the time, they made sense. We were building for a world of linear, predefined workflows. Digital Transformation was about creating faster structured journeys. It was about automating the "if-this-then-that" logic of traditional banking.
AI is not the same. Trying to apply the same principles won't work.
It is a fundamental error. We aren't managing a "program" anymore: we are managing intelligence. If you try to treat a non-deterministic AI teammate like a rigid piece of software, you will stall.
The Autonomy Paradox
In the old Digital Transformation world, you could test a journey until you had 100% certainty of the outcome. With AI, that certainty doesn't exist in a lab.
Operational leaders want the productivity gains of autonomous agents, but Risk and Compliance teams are (rightly) terrified of the "Black Box." If you grant an agent autonomy before it has earned trust, you create unmanaged operational risk. But if you wait for "100% certainty" through traditional 12-month testing cycles, you never deploy.
The result?
Projects die in the lab because the risk of "unearned autonomy" is simply too high for a CTO or CRO to sign off on.
Moving to Hybrid Controls
Next week at the MoneyLIVE Summit, I’ll be joining a fireside chat to discuss how we break this deadlock through Scaling AI Safely.
The solution isn't to build a better "test." The solution is to change the architecture of the deployment using Hybrid Controls, where Human and AI Teammates do their fair share in ensuring quality and compliance.
We have to move away from retrospective auditing. We need to stop checking what went wrong six months ago and move toward "Guardrails by Design." This means engineering the controls into the model from day one so that policies are enforced in real-time.
Trust First, Integration Later
The biggest hurdle to AI adoption isn't the technology: it’s the "Integration-First" mindset.
Most firms believe they have to get data access to AI and integrate into data stores or even core system before they can see value. This is a recipe for a two-year roadmap that delivers zero immediate benefit.
At Vigilant AI, we advocate for a different path that emulates the operational onboarding of junior hires...
Onboard, Collaborate, Trust then Scale.
1. Onboard: Treat the AI Teammate like a new hire. Give it your training, your specific policies, your processes, and teach it your tone of voice.
2. Collaborate: Embed the teammate directly into your collaboration environment. Don't start with full autonomy: start with "Human-in-the-loop" oversight, and provide coaching and feedback to help it improve.
3. Trust: Let the teammate earn trust through performance, monitored by a real-time Supervisor layer that ensures compliance with your overarching policies and regulations.
4. Scale: Once you've built trust then, and only then, look at potential integrations and scaling autonomy.
This way, once the trust is proven and the audit trail is visible to the entire team, you can begin the heavy lift of deep system integration. You get value, you build the logs, and you reimagine the processes without the upfront effort and risk of failure.
Start in Operations, Not IT
We have to stop thinking about AI as "Automation 2.0."
Automation is about doing the same simple repeatable task 1,000 times. AI is about doing the difficult, complex reasoning work that requires real thought.
The first you do in IT, because you can define, test it and fully prove it. The 2nd one is non-deterministic so you have to do it within you operation. With our operational expert SMEs who can provide oversight and accept that there is no "100% correct" answer.
The successful banks of 2026 won't be the ones with the longest roadmaps. They will be the ones that understand that AI doesn't need another two-year project plan, but its a process of continuous improvement.
It needs a reporting line
It needs to be managed, governed, and coached just like any other high-performing intelligence in your business. I look forward to sharing the practical evidence of how this is already working next week.
March 6, 2026
Michael Anyfantakis, Co-Founder
March 6, 2026
I was there for the first wave of Digital Transformation in banking.
I’ve seen the 3 year roadmaps and 12 month plans, the exhaustive requirements gathering, and the extensive testing cycles. At the time, they made sense. We were building for a world of linear, predefined workflows. Digital Transformation was about creating faster structured journeys. It was about automating the "if-this-then-that" logic of traditional banking.
AI is not the same. Trying to apply the same principles won't work.
It is a fundamental error. We aren't managing a "program" anymore: we are managing intelligence. If you try to treat a non-deterministic AI teammate like a rigid piece of software, you will stall.
The Autonomy Paradox
In the old Digital Transformation world, you could test a journey until you had 100% certainty of the outcome. With AI, that certainty doesn't exist in a lab.
Operational leaders want the productivity gains of autonomous agents, but Risk and Compliance teams are (rightly) terrified of the "Black Box." If you grant an agent autonomy before it has earned trust, you create unmanaged operational risk. But if you wait for "100% certainty" through traditional 12-month testing cycles, you never deploy.
The result?
Projects die in the lab because the risk of "unearned autonomy" is simply too high for a CTO or CRO to sign off on.
Moving to Hybrid Controls
Next week at the MoneyLIVE Summit, I’ll be joining a fireside chat to discuss how we break this deadlock through Scaling AI Safely.
The solution isn't to build a better "test." The solution is to change the architecture of the deployment using Hybrid Controls, where Human and AI Teammates do their fair share in ensuring quality and compliance.
We have to move away from retrospective auditing. We need to stop checking what went wrong six months ago and move toward "Guardrails by Design." This means engineering the controls into the model from day one so that policies are enforced in real-time.
Trust First, Integration Later
The biggest hurdle to AI adoption isn't the technology: it’s the "Integration-First" mindset.
Most firms believe they have to get data access to AI and integrate into data stores or even core system before they can see value. This is a recipe for a two-year roadmap that delivers zero immediate benefit.
At Vigilant AI, we advocate for a different path that emulates the operational onboarding of junior hires...
Onboard, Collaborate, Trust then Scale.
1. Onboard: Treat the AI Teammate like a new hire. Give it your training, your specific policies, your processes, and teach it your tone of voice.
2. Collaborate: Embed the teammate directly into your collaboration environment. Don't start with full autonomy: start with "Human-in-the-loop" oversight, and provide coaching and feedback to help it improve.
3. Trust: Let the teammate earn trust through performance, monitored by a real-time Supervisor layer that ensures compliance with your overarching policies and regulations.
4. Scale: Once you've built trust then, and only then, look at potential integrations and scaling autonomy.
This way, once the trust is proven and the audit trail is visible to the entire team, you can begin the heavy lift of deep system integration. You get value, you build the logs, and you reimagine the processes without the upfront effort and risk of failure.
Start in Operations, Not IT
We have to stop thinking about AI as "Automation 2.0."
Automation is about doing the same simple repeatable task 1,000 times. AI is about doing the difficult, complex reasoning work that requires real thought.
The first you do in IT, because you can define, test it and fully prove it. The 2nd one is non-deterministic so you have to do it within you operation. With our operational expert SMEs who can provide oversight and accept that there is no "100% correct" answer.
The successful banks of 2026 won't be the ones with the longest roadmaps. They will be the ones that understand that AI doesn't need another two-year project plan, but its a process of continuous improvement.
It needs a reporting line
It needs to be managed, governed, and coached just like any other high-performing intelligence in your business. I look forward to sharing the practical evidence of how this is already working next week.
