By Igor Nikitin, ASA, MAAA
CEO, Nice Technologies LLC
One of the biggest misconceptions about AI-assisted software development is that generating code is the same thing as delivering software. Just because AI wrote it in a day doesn’t mean it is production-ready in a day.
Before AI, you might spend a year building a system and understand every corner of it by the time it was deployed. Today, AI can help generate a significant portion of a system in a matter of weeks. However, generating the code is often just the beginning of the process.
The bottleneck is no longer writing code. The bottleneck is gaining confidence in the code. In insurance, especially, time-to-production matters more than time-to-generation.
Once AI generates the code, the work is just beginning:
- Verifying that there are no mathematical mistakes
- Verifying maintainability, including architecture, code clarity, and performance
- Making sure someone understands the codebase if something stops working
- Satisfying governance and audit requirements
- Verifying security
- Ultimately, getting the chief actuary and other stakeholders comfortable enough to rely on the results
AI can generate thousands of lines of code faster than a team can realistically review and understand them. In regulated industries, the challenge is often not building the software. It is proving that the software is correct, maintainable, secure, and trustworthy.
Despite these challenges, AI remains one of the most important productivity advances we’ve seen in software development. The key is understanding where it accelerates the process and where human expertise remains essential.
For insurance companies looking to move faster, here are three practical ideas:
1. Bring in knowledgeable leadership
AI adoption needs a champion who understands both the technology and the business. You need someone who can separate hype from value, identify realistic opportunities, and understand the realities of insurance organizations.
2. Don’t reinvent the wheel
Before building internally, evaluate whether proven commercial solutions already exist. Many organizations underestimate the build investment, long-term maintenance, governance, and support costs associated with custom development. In many cases, an off-the-shelf solution can deliver value faster and at a lower total cost.
3. Give employees a safe place to experiment
Provide employees with sandbox environments and time to explore AI tools. Just as importantly, listen to what they discover. Some of the most valuable AI use cases come from people closest to the day-to-day work. Organizations that systematically identify and scale successful experiments will move faster than those relying solely on top-down initiatives.
AI is changing how software gets built, but generating code is only one step in the journey. The organizations that gain the most from AI will be the ones that can validate, govern, and deploy that code with confidence.