INTERVIEW
What's next for AI in banking?
An interview with the winner of MoneyLIVE North America's Startup Award, Cobalt
1. For those who are new to Cobalt, tell us what you do - what problem is Cobalt solving?
Over the past two years we’ve spoken with more than a hundred financial institutions about their AI transformations, and the same pattern comes up everywhere.
AI has genuinely solved coding. Practically every bank has put Copilot or a similar assistant in front of every engineer, the models are excellent, and the cost of writing a new line of code has fallen to almost nothing. Greenfield work and prototyping have never been faster.
What AI has not solved is brownfield – the systems that already exist and that everything else depends on. And the reason is that in brownfield, coding was never the hard part. The hard part is deriving the specification for the change, and then validating the result to the standard a bank requires. Neither of those has gotten meaningfully faster.
In one bank we work with, every developer has Copilot and has had it for months. They are still afraid to touch a legacy mortgage system written in COBOL, because nobody properly understands how it works.
That is the gap Cobalt exists to close. In a typical bank, a single ATM withdrawal traverses around twenty systems. Business logic sits in mainframe batch jobs, database stored procedures and opaque third-party applications. Documentation is stale or missing, and many of the people who built it are no longer around to ask. None of that is a criticism – it is what forty years of successful banking looks like from the inside. But an agent cannot read it, so it guesses, and a guess in a payments path is not a productivity gain, it is an incident.
Cobalt maps the bank’s systems and business processes and gives that map to the agents as context. Our goal is to make brownfield development as efficient as greenfield.
2. What sets Cobalt apart? What's your "secret sauce", and why can you win in this market?
My short version of the whole build-versus-buy debate is: buy the tooling, rent the models, but own the map of your bank.
Code used to be the IP. It isn’t any more – code can be generated, and it is really just a temporary expression of a specification. What matters now is the specification. And in a bank, getting an accurate specification for any system of consequence is harder than anywhere else, because of the business processes, the thousands of components, the mix of proprietary systems, third-party software and cloud services, and the countless hidden dependencies. So agents in a bank need more than a spec. They need a map of how everything is connected.
That map is what we build. Our engine, which we call the Semantic Crawler, reads across all the sources that together explain how the bank actually works – code, logs, telemetry, CMDB and documentation – and traces the connections between business processes and the IT assets that run them, in both directions.
Three things make it different. Nobody draws it: it is inferred from code and logs, so it reflects what is genuinely running rather than what someone documented years ago, and it re-crawls continuously, so the map moves when the estate moves. Your CMDB tells you what exists; Cobalt tells you what it does and how it’s connected. Second, it is built to be consumed by agents, not only by humans – the same context that lets an architect ask “what might break if we change this limit” is served directly to the coding agents a bank already licenses, and every answer comes back grounded with an evidence trail. And it runs read-only, inside the bank’s own perimeter; no source code leaves the environment.
The map is also the one asset in the AI stack that appreciates rather than depreciates. Models will be replaced, tools will be replaced. What everything else gets built on is knowing how your own bank works.
3. What's next for Cobalt? What are you most excited about as you look ahead?
Two things.
The first is watching coding agents graduate from assistants to executing entire stages of the development lifecycle – design, implementation, verification, deployment. One of our design partners, a publicly traded trading platform, is rewriting an entire customer-facing system built on hundreds of repositories with no documentation anyone trusts. The conventional path for that is about two years, which is why it stalled for years. With Cobalt mapping the system – including inferring from logs what is genuinely running, not just what is coded – they have kicked it off as an agent-led rewrite: agents write the code, humans own the review gates. It is scoped at three months.
The second is everything beyond development. Another design partner uses the same map for operational resilience. When an incident hits the operations center, the first hour almost never goes to fixing it – it goes to working out which system is actually responsible, and the people who hold those dependencies in their heads are not necessarily the people on call. Their SRE agent queries Cobalt live during the incident and gets an evidence-backed root cause in minutes, with the trail that proves it. Vendor swaps, modernization, incident response – once the map exists, all of it becomes automatable.
What I’m most excited about is where that leads organizationally. Banks today have a head of AI. I think within a couple of years they’ll have something closer to a head of operational truth – someone accountable for whether the bank’s description of itself is still accurate, the way a data steward is accountable for lineage. That is the layer everything else now depends on.
Oren Buskila is Co-Founder and CEO of Cobalt, a company helping banks and financial institutions understand the real structure of their technology environments through continuous discovery and operational intelligence. Prior to Cobalt, Oren held senior technology and product leadership roles in large scale software and enterprise technology organizations, where he focused on complex systems, digital transformation, and building products for highly regulated industries. At Cobalt, he works closely with leading financial institutions to help them accelerate modernization initiatives, reduce transformation risk, and prepare their organizations for AI adoption.
Cobalt is purpose built for financial institutions. By continuously mapping enterprise systems, applications, data flows, and dependencies, Cobalt gives AI the understanding it needs to accelerate software delivery, reduce operational risk, and modernize software engineering across the SDLC.
Hungry for more? Check out
MoneyLIVE North America
Join 900+ senior leaders at North America's most senior banking and payments event
Have a story to share?
Book a call with us to learn more.
Hannah Johnson Business Development Manager MoneyLIVE
Stay in the loop with MoneyLIVE
Subscribe to our newsletter to receive news, insights and special offers.