From a company perspective, what represents the most valuable knowledge an engineering team possesses: Experience with specific frameworks and languages, Mastery of development tools? The answer to both questions is no. The most critical knowledge lies in a deep understanding of the company applications and business domain. This expertise enables teams to interpret requirements accurately, address unstated assumptions, and implement solutions effectively within the existing codebase.

Building this understanding takes time, especially as much of this knowledge does not exist in formal documentation. Much of it represents tacit knowledge accumulated through daily problem-solving, evolving business rules, and subtle system interactions that change rapidly in high-growth environments. Formal documentation rarely keeps pace with these realities and seldom captures assumptions, edge cases, or the nuances that only emerge through prolonged exposure to the actual codebase and operations.

This has always been a challenge when growing teams rapidly. New engineers need to gain this knowledge from the existing staff. If you bring on a lot of new people in a short amount of time, your existing engineers will spend a lot of time coaching and less time developing. Hiring new people can, initially, at least, reduce productivity.

This problem is even more pronounced when the new engineer is an AI agent. Not only does an AI agent share the same lack of context with a new hire, but it is not as good at learning as a human, especially when that information is not documented.

I have seen first-hand that AI speeds up converting requirements into code. However, AI-generated code takes longer to pass a PR review. There is still a net gain, but a less impressive one than if you focused on the coding alone.

Companies who have been aggressively terminating engineers because of AI may well find themselves struggling to deliver in the medium term. AI needs the guidance of engineers who have institutional knowledge. Without it, AI will continue to deliver code, but with an increasing likelihood of defects and technical debt. This is not a problem that can be fixed by hiring, since those new people also lack that knowledge.

Measuring software engineer productivity has always been difficult. Evaluating gains from AI agents is no easier. Betting a company future on AI while removing the people needed to make it succeed is short-term thinking. Companies firing software engineers are discarding some of their most important assets. AI productivity gains may not offset that loss.

#InstitutionalKnowledge #SoftwareEngineering #AITechnology #TechLeadership #EngineeringProductivity