The pace of hardware growth is constantly accelerating, and with it an ever-growing list of hardware security requirements. As a leader in the hardware security space, we are obsessed with continuously delivering cutting-edge products for our customers to enable them to build trust at the chip level. With the recent surge in AI-driven development products, we recognize the value these tools provide when integrated properly into our engineering workflows and have, like many engineering organizations, adopted this into various parts of our day-to-day lives.
Our Approach
Integrating artificial intelligence into different facets of our development process requires careful planning and integration to avoid the pitfalls and tribulations that come from improper overuse (looking at you – “vibecoding”). We’ve seen an overwhelmingly positive net impact on our day-to-day effectiveness, with the team citing key benefits like simplifying tedious tasks and quicker feature implementation. Therefore, it was important for us to treat AI-assisted development as more of a virtual engineer rather than an autonomous contributor.
What this means for us is treating AI as yet another tool in our toolbelt, using it primarily for feedback on design, implementation, and debug—the long-pole steps in our development process—which can succumb to human error. The success of this disciplined approach is clearly reflected in the key findings from our recent engineering team poll, which provide insight into how AI is enhancing our daily workflow:
Key Findings from Our Engineering Team
- Productivity Boost: A qualitative assessment of our team’s use of AI showed the overall impact on productivity was unanimously positive, with a clear consensus on a strong net positive for day-to-day effectiveness.
- Top Benefits: The biggest benefit is in tasks like writing boilerplate or repetitive code, but it also significantly shortens the time to prototype, quickens feature implementation, and helps provide valuable knowledge of existing APIs or libraries.
- Quality Impact: We are seeing a high impact on code quality, with a majority of the team citing that AI makes both code correctness and bug frequency better.
- Known Downsides: While the impact is positive, downsides mentioned were related to AI at times generating incorrect responses or going too far off course, as well as creating overly complex changes. Sometimes, overconservative feedback in the form of “nitpicks” can slow down the code review process if the AI is too aggressive. Many of these qualitative issues are expected to become less prominent as models evolve and mature over time.
Paired Programming Reinvented
However, even with these minor drawbacks, the benefits are clear, particularly when AI acts as an instant collaborator, reinventing the traditional paired programming model. Traditionally, companies adopted a “pair programming” working model, which required the time investment of two engineers for bounded ranges of time to reduce risk and improve code quality. While costs versus benefits for this practice are still hotly debated, the use of AI in our development workflow essentially gives us the benefits pair programming provides, with seamless instant integration and without the need to have multiple engineers working on a single issue. Integrated tab completion fixes small errors, reducing the time spent debugging both in compile and potentially in the field, while chat panels and coding agents make bouncing and implementing ideas more of a conversational interacting than rudimentary work.
Improving Code Quality
The most high-risk portions of the development process, from coding to review and deployment, usually suffer from human error. Developers have all had a time where a small typo, using the wrong variable, or forgetting to set a configuration results in hours of rework or worse, a critical bug at deployment. To help reduce this human error, we added AI to our Pull Request process to provide another set of eyes on code being introduced to our main git branches. While IDEs, static code checkers, and various CI/CD tools have helped to mitigate this, adding AI into our CI/CD workflow at the pull request stage has caught implementation issues, design considerations, and potential bugs that other tools have more difficulty finding. In fact, our engineering team’s poll revealed that using AI has made code correctness and bug frequency “Better,” validating this approach to reducing human error.
For example, the context awareness of AI-integrated pull requests has caught issues and flagged considerations ranging from language-specific string handling, feedback on patterns from widely used languages like SQL, and flagging expression evaluation issues which may lead to unintended behavior. Many of these are typically not flagged by the tools we use today simply because finding these types of issues requires contextual rather than purely structural knowledge – something AI is now providing.
Closing the Debug Loop
Every software company faces this issue: bugs. They happen, and when they occur the time to recover can be crucial for the customer and the vendor. For vendors like us who work with companies producing hardware, access can be a sometimes insurmountable hurdle, which causes delay for the customer receiving a fix.
AI-assisted development has given us the ability to better understand and triage problems as they come in- both faster and with more contextual understanding. For example, given a bug scenario, an engineer can easily transform the scenario into a prompt description, add context of the software systems which may be involved, and have AI develop a debug plan or, in many instances, isolate the problematic areas for you.
In a recent instance, AI-driven development allowed us to debug a corner-case issue experienced in the field with no access to an environment or design. Based on pure symptom description, we were able to identify and patch the issue in minutes. Specifically, this bug was dormant in a piece of code which had existed for so long it was practically legacy, and the issue was so benign it evaded human eyes as well as our CI/CD code checks. In a traditional workflow, maybe one or two years ago, this would have been a game of telephone lasting hours or days, with an unbounded amount of time to issue a patch at great cost. However, with AI assisted workflows, triaging and diagnosing issues is becoming an easier task, closing the gap between issue discovery and resolution.
Conclusion
AI is providing our engineers with new ways to amplify their potential. By treating AI as a disciplined collaborator, one that excels at context, pattern recognition, and rapid feedback, we’ve reduced risk, improved code quality, and tightened the debug loop without sacrificing rigor. The result is faster, more reliable delivery of hardware security tools our customers can trust, and a development process that keeps pace with the speed of modern hardware.