Alister Mac Cormack: AI Can Write Code And Engineers Need to Understand It

The forward-deployed engineer says faster code generation increases the need for people who can direct AI systems, review their work, and understand the problems they are being asked to solve.

Alister Mac Cormack hears frequent claims that artificial intelligence has reduced the value of learning to code. His work inside companies keeps showing him why technical understanding still matters. AI can produce software quickly, but speed does not determine whether the approach is correct or whether the result belongs inside the system where it will be used.

“The tool can produce code quickly, but someone still has to know whether the pattern is correct and whether the result works,” Mac Cormack said. “You need enough engineering knowledge to give it useful direction and catch the mistakes.”

Mac Cormack works from Chile as a forward-deployed engineer at Cognition, where he was the first person hired into the position from Chile. He studies how a company operates and learns its technical systems in depth before deciding where AI may be useful. Some assignments involve adapting an existing product. Others require custom engineering because the standard option leaves important gaps.

Mac Cormack sees a meaningful difference between asking an AI system to generate code and guiding it toward a dependable result.

The first task can begin with a written instruction. The second requires an engineer who understands the surrounding system and can recognize the correct technical pattern. That person must also examine the output closely enough to know when it only appears convincing.

“If you cannot think like an engineer, it becomes difficult to give the agent good direction,” he said. “The answer may look right even when the approach behind it is wrong.”

An AI-generated response may run successfully while solving only part of the problem. It can also follow a familiar approach that does not fit the company’s actual systems.Mac Cormack begins by understanding what employees are trying to accomplish and where the current process breaks down. That investigation helps him decide whether an available AI product can be adapted or whether a custom solution is necessary.

“I need to understand the company’s systems and the problem in depth before I can decide how AI should be used,” he said. “You can build something that works and still miss the problem people needed you to solve.”

The engineer’s responsibility extends beyond producing software. Mac Cormack must judge whether the proposed solution belongs inside the business and whether it can function reliably within the existing environment.

A demonstration can show that an AI product is capable of completing a defined task. Mac Cormack’s work continues after that initial proof.

A company has its own systems and operating problems. A standard product may address part of the need while leaving important gaps. Engineers must identify those gaps before employees begin relying on the technology.

“Something can work well in a demo and still need a great deal of engineering before people can depend on it,” Mac Cormack said. “The important part is making it useful inside the company rather than proving that it can complete the task once.”

Coding knowledge remains valuable during that transition because the engineer must understand how the new work interacts with software already in place. Generated code needs to be reviewed in context rather than accepted because it arrived quickly.

Mac Cormack examines whether the output reflects the right pattern and fits the problem he was asked to solve. He must also recognize when the tool has made an assumption that does not hold inside the company.

The review may lead him to revise the instruction, change the approach, or discard the generated response.“You cannot treat the first answer as the finished solution,” he said.

“You have to keep checking the work against the system and the result the company needs.”

His cybersecurity background also makes him attentive to risks that generated code may introduce elsewhere, even when the immediate task appears to work.

“You have to look past the first result and ask what else the change affects,” Mac Cormack said. “That is part of understanding whether the solution is actually sound.”

Mac Cormack encourages engineers to gain exposure to different industries and technical problems. His own work has crossed fields as different as medical systems, residential solar, and cybersecurity.

He believes that range gives engineers more ways to examine whether an AI-generated answer fits the problem in front of them. Experience from one setting can reveal a weakness or possibility in another.

“The useful connections often come from experience that looks scattered at first,” he said. “You recognize something because you have seen a version of the problem somewhere else.”

Broad exposure gives Mac Cormack more reference points, while technical depth allows him to test whether an apparent connection holds up. Forward-deployed engineering requires both because each company presents a different operating environment.

The answer cannot be based only on what the AI is capable of generating. It must also reflect what the company can realistically use.

The speed of AI-generated code can create the impression that engineers no longer need years of practice. Mac Cormack believes practice remains the foundation of reliable review.

Engineers build judgment by writing software and debugging failures. They learn from unfamiliar code and from solutions that seemed reasonable before breaking under real conditions.

“There is no shortcut around the hours,” he said. “Working smart matters, but the repetitions are what teach you how to recognize a good solution.”

AI can help engineers attempt more work within those hours. It can produce a starting point and reduce the time spent on routine tasks. The engineer still has to decide whether that starting point is useful.

Mac Cormack sees increasingly capable tools as an opportunity to tackle harder problems. Technical knowledge helps an engineer provide clearer instructions and recognize sooner when the generated answer should be revised.

“As the tools improve, you still need to understand enough to check their work,” he said. “Otherwise, you are relying on an answer you may not be able to evaluate.”

The daily work of software engineering may continue to change as AI systems become more capable. Mac Cormack expects engineers to spend less time producing every line manually, with more attention shifting toward directing the technology and reviewing what it produces.

Responsibility still belongs to the person deploying the work. An AI system does not understand a company simply because it receives technical instructions. Someone must determine which result fits the organization and which compromise is acceptable.

Those decisions require an engineer who understands the technology and the setting where it will operate. The ability to generate code more quickly does not remove that obligation.

For Alister Mac Cormack, code appearing on a screen marks the beginning of the review. The engineer still has to decide whether the result belongs inside a system people will depend on.