Council Post: Human-In-The-Loop Is No Longer Enough: AI-In-The-Loop Is The Future
Dr. Son Nguyen is Orient Software’s CTO & co-founded Neurond AI, companies specializing in software development, AI, & data science servicesgettyA couple of years ago, whenever our engineers used AI in...
Dr. Son Nguyen is Orient Software’s CTO & co-founded Neurond AI, companies specializing in software development, AI, & data science services
getty
A couple of years ago, whenever our engineers used AI in software development, I always gave the same advice: "Keep yourself in the loop." Generate the code with AI, review every change, verify every implementation, and never let the model run unsupervised.
At the time, that was the right approach. AI was a capable coding assistant, but it still made enough mistakes that every iteration required human oversight.
Watching how quickly technology has evolved across the industry and inside our projects, I believe that mindset is becoming outdated. Modern models can plan, implement, test, identify failures and refine their work through multiple iterations until they reach the desired outcome.
This has led me to rethink a long-held assumption: Does every workflow still need a human in the loop? Increasingly, I believe the answer is no. We need to redefine the role of developers, from supervising steps to setting objectives and success criteria while allowing AI to execute and iterate autonomously.
The shift from human-in-the-loop to AI-in-the-loop is a fundamental change in how we build software. I’ll explain why this shift is happening and how engineering teams can take advantage of it.
What AI-In-The-Loop Is
AI-in-the-loop (or loop engineering) refers to a workflow in which an AI system continuously works toward a goal with minimal human intervention. Instead of generating a response and waiting for the next prompt, the AI iterates through a series of steps, evaluating and improving its own work until it achieves the desired outcome.
The process typically begins with a developer defining a high-level objective, such as implementing a new feature or fixing a bug. The AI then plans, writes code, runs tests, checks the results and iterates until it meets the success criteria or reaches a stopping condition.
What makes this approach possible is the rapid advancement of AI capabilities. Modern models can reason through complex problems, use tools, analyze results and improve their own solutions.
How AI-In-The-Loop Differs From Human-In-The-Loop
The answer lies in who's driving it. In other words, it's how humans and AI share responsibilities.
Human-in-the-loop lets developers guide every iteration. AI acts as an assistant by generating code, suggesting improvements or debugging issues based on the developer's prompts, while humans review the output, provide feedback and decide the next step. In contrast, AI-in-the-loop shifts much of the execution to the AI, forcing it to plan, implement, test and refine its solution until the goal is achieved.
Neither approach is better than the other. Human-in-the-loop excels for tasks that require creativity, product judgment or frequent decision making. Meanwhile, AI-in-the-loop is ideal for repetitive, well-defined tasks, enabling developers to focus on higher-value engineering work.
Why AI-In-The-Loop Is The Future
Its growing interest is due to the ability to enable AI to solve problems through continuous iteration rather than one-time responses. The feedback loop often leads to more reliable and refined results. This also frees developers from supervising every implementation step, allowing them to focus on work such as system design, complex problem-solving and product decisions.
AI-in-the-loop makes better use of the model's reasoning capabilities. Because the AI can explore multiple approaches, learn from failed attempts and refine its solutions, it has more opportunities to produce optimized or creative results than in a traditional prompt-and-response workflow.
While the technology is still evolving, AI is clearly moving from being a coding assistant to becoming an autonomous collaborator.
Where AI-In-The-Loop Actually Delivers Values
Across coding, mathematical reasoning and medical diagnosis support, frontier AI models now match or even surpass human performance on many benchmarks.
One common application is automated testing. AI can generate test cases and test suites, analyze failures and modify code until all tests pass. Similarly, it can refactor repetitive code and improve readability and maintainability while ensuring existing functionality remains intact.
AI loops are also effective for fixing linting issues, updating documentation and upgrading dependencies. As these tasks follow established rules and can be validated automatically, they're well-suited to autonomous iteration.
Beyond code changes, AI-in-the-loop can support CI/CD pipelines, security scanning and code reviews. It identifies vulnerabilities, suggests fixes, reruns validation checks and continues refining its output until predefined quality standards are met.
Challenges Teams Should Consider
Despite its potential, AI-in-the-loop isn't a one-size-fits-all solution, and teams should understand the trade-offs before adopting it.
First is the cost, as it consumes significantly more AI tokens than traditional prompt-response interactions. For long-running or complex tasks, these costs can quickly add up.
Another challenge is the assumption trap. AI models are designed to fill in missing information to keep making progress. When requirements are unclear, AI may make incorrect assumptions that compound over multiple iterations. Providing clear objectives, constraints and evaluation criteria helps reduce this risk.
Finally, not every aspect of software development can, or should, be automated. Decisions about user experience, product strategy or brand voice often require human judgment. AI can generate and evaluate solutions, but humans remain responsible for choosing the one that delivers the greatest value.
Conclusion
AI-in-the-loop marks an exciting step toward more autonomous software development. The future is about knowing when to let AI take the lead. Teams that embrace this shift will be better positioned to increase productivity and build software more effectively.
As it continues to mature, now is the time to experiment with agentic workflows and rethink how humans and AI collaborate.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?