Sapient Slingshot wins Technology Breakthrough of the Year.
Sapient Bodhi ranked #1 for deep research.
Meet Sapient Sustain: context-aware AI for complex IT operations.
3M lines of COBOL → clear specs in 8 weeks.
Building AI agents for a commercial banking proof of concept revealed that enterprise AI works best when every agent has a clear purpose, the system is grounded in trusted information and human judgment remains central to the final decision.
September 16, 2026
Loading AI-generated summary...
My AI journey at Publicis Sapient began with a practical question: how can AI help solve complex problems that large enterprises care about?
For me, that question was closely connected to financial services, especially commercial banking. Having spent years in banking technology, I knew enterprise problems involve systems, users, rules, documents, risks and decisions that depend on trusted information.
That shaped my approach. I was not looking for a use case just because AI was available. I was looking for places where AI could help people work with information more effectively.
One early area I explored was commercial onboarding. When a new business customer joins a bank's platform, the process can involve long forms, documents and manual review. My team and I looked at how AI could extract information and make the process easier.
Another use case pushed my thinking further: supporting relationship managers in commercial banking.
Their role requires them to understand financial health, industry context, risks, compliance needs, growth potential and relevant products or services. This research can take days or weeks.
I asked a practical question: could AI help relationship managers move faster while still giving them useful reasoning behind every recommendation?
That question led to a proof of concept with more than 50 AI agents—specialized software components designed to perform specific tasks, analyze information and pass their findings to other agents within a coordinated workflow. Each agent focused on a specific part of the problem, such as financial health, sector outlook, risk, compliance or brand performance.
My goal was not simply to collect information. It was to bring different signals together and create a clearer synthesis for better decisions.
For me, this distinction mattered. AI was not replacing the relationship manager's judgment. It was reducing the time spent searching, gathering and organizing information, so people could focus on the parts of the role that require experience, context and client understanding.
The work also required discipline. Not every task needs AI. Some problems can be solved with automation, where the work is predictable and rules-based. AI becomes more useful when the task involves reasoning, analysis, synthesis and judgment.
That helped me and my team stay focused. We did not start with the tool. We started with the business problem and asked where AI could add value.
Public information helped with the proof of concept. Deeper enterprise value would come from combining it with private data, such as transactions, customer relationships, service requests, product information and internal knowledge graphs.
That is where the work moved beyond a demo and became a real enterprise AI challenge.
Building a proof of concept with more than 50 AI agents required more than individual experimentation. I needed access to the right tools, people and technical support to test ideas quickly and improve the system as I learned.
Publicis Sapient's agentic AI platform, Sapient Bodhi, helped reduce the setup work involved in creating, deploying and managing agents. Instead of building the underlying environment from scratch, I could focus on the commercial banking use case itself—testing workflows, comparing models, reviewing logs and debugging issues.
The wider network mattered too. Colleagues with experience in AI, engineering and financial services gave me different perspectives on the problem. That combination of platform support, domain knowledge and technical collaboration helped me move from an early idea to a more structured and credible proof of concept.
At one point, I realized that a single large prompt was not the best approach. It consumed more tokens, increased cost and made the workflow harder to manage. So I applied a familiar software architecture principle: single responsibility.
Instead of one large agent trying to do too much, I broke the solution into smaller, modular agents. Each had a clear role, making the workflow easier to test, reuse and adapt. A component I built for relationship management could also support brand health analysis or risk review.
That is the kind of thinking enterprise AI needs: curiosity about new tools, but also disciplined design.
In financial services, trust is essential. Users need confidence in the insights they see, especially when those insights may influence business decisions.
My team and I worked on ways to make the workflow more reliable. We bounded the input and output context between agents and carried confidence scores through the workflow, so recommendations could show how strongly they were supported by the available data. A higher score indicated that the evidence was clearer and more consistent, while a lower score signaled that the recommendation needed closer human review. This gave users a simple way to understand where the AI was more certain and where judgment was still required.
This helped make AI more usable in a real business environment.
For me, the expected impact is clear. AI can move work from days or weeks to hours. It can reduce time spent collecting and analyzing information and help users make decisions with better context.
At Publicis Sapient, I found a culture where people are encouraged to learn, try and apply new skills to real problems. Leaders support experimentation, teams bring different strengths and opportunities come through client work, RFPs, prototypes and proofs of concept.
For AI talent, that is meaningful. The future of AI will be shaped not only by people who understand models, but also by people who understand industries, systems, users and business decisions.
My journey shows what that can look like: AI tested against real problems, shaped by domain knowledge, built with engineering discipline and supported by teams that help ideas become useful enterprise solutions.
Explore our current opportunities.
