[3/4] Panel discussion on the impact of AI on software development (AI column)
Continuing the story (1 and 2) I will share the remaining topics and start with the risks.
3. Opportunities vs. Risks: How to Balance Acceleration and Threats The advantages of AI in development are obvious and enticing. The main drivers that companies call: increased productivity, faster releases and reduced costs. But it carries a trail of risks. - Quality and error risks. AI tends to “hallucinate” – give confident but incorrect answers. Developers complain that it takes a lot of time to debug the code, which is almost working. The strategy here may be to create additional checks. Autotests, static analysis, linters are mandatory even for generated code. (Especially for him!). Some companies go a step further, requiring them to tag or revise all code entered with AI. Risks to the organization and processes. Without a competent approach, AI can strengthen bottlenecks. For example, DORA warns that if a company suffers from outdated processes, technical debt, lack of automation, then the acceleration of development through AI will lead to an avalanche of problems – more raw code will flow into the same pipeline curves. In this case, in addition to the introduction of AI tools We need to develop a development platform and engineering processes. - Data privacy and security. AI tools often require sending source code or data to external services. For financial institutions, this is a red flag. Such companies deploy AI solutions locally and refine them to suit their own scenarios. If you have a contractual relationship with LLM model providers, then you can try to solve this problem at the contract level. - Vendor lock-in. Now the AI platform market is mostly controlled by a few players – OpenAI (Microsoft)Google, Anthropic, etc. If a team is tightly attached to, say, GitHub Copilot, there is an addiction, so large firms try to diversify and keep control. The practice of large companies is to develop internal expertise and/or use the services of different providers. (multiplier) Competences of the team and “atrophy of skills”. Ironically, we want the team to work faster with AI, but we’re afraid that it will stop people growing as professionals. Here everything is tied to training employees and motivating them not just to copy the answers of the model, but to deal with them and understand how and why it works.
4. How do development processes and team composition change? The widespread adoption of AI is already beginning to change the structure and roles in development teams. In the coming years, the need for testers and analysts will decrease, and the number of development engineers in the state will increase. - Why fewer testers? Traditional manual testing is one of the processes most amenable to AI automation. Modern tools are able to generate test scenarios themselves, select edge cases and even keep tests relevant when changing the code. - Why fewer analysts? The word analytics here can be understood as business analysts, system analysts – those who translate business requirements into specifications. The AI began to eat away at this part of the work.
Already today, the impact of AI is being felt in team roles: developers become the center of the process, relying on AI, they take on part of the testing and analytics tasks; testers and analysts are being transformed into more technical and combined roles, and new positions are emerging. For bigtech and fintech, this is a chance to optimize team structures and increase efficiency, but success depends on retraining people. The human role will shift to solving non-standard problems, making product decisions, and guiding AI, while AI will take on more routine. Teams will become smarter and more cross-functional, and people will become more versatile.
The ending will be in the final post.
#AI #Engineering #Architecture #ML #Software #Economics #Software