[4/4] Panel discussion on the impact of AI on software development (AI column)
Finishing the story (1, 2 and 3) I’ll talk about the last two topics.
5. Is there trust in AI capabilities or is it just hype? Despite all the successes, the industry feels ambiguous attitude to AI tools. On the one hand, everyone is talking publicly about AI-first strategies and how neural networks affect efficiency. On the other hand, on the sidelines, many engineers and managers admit that they do not fully trust these solutions and sometimes “pretend to support the trend”, adjusting to fashion. Where is the line between real belief in technology and imitation of enthusiasm? Objectively, the level of trust has decreased compared to the excitement of the first years. According to surveys, the share of developers with a clear positive attitude to AI tools in the 2025 fell before ~60%, whereas in 2023 was> 70%. That is, the initial excitement cooled down, the understanding of limitations came. DORA Report speaking The paradox of trust: although 90% use AI, only 4They fully trust him and only 20The percentage trusts "significantly." Around 30The percentage admits that there is little or no trust. In other words, almost everyone finds AI useful, but almost no one finds it unmistakable. As a result, now many follow the saying “trust, but verify” – for high-risk operations still need human in the loop, but less risky operations are trying to entrust conditional AI-agents with possible post-control. (Review design, review code, test phase). People are still an integral part of the process.
6. What awaits development on the horizon of 3 years? Will the singularity come or will the hype collapse? Looking at 3 It is safe to say that AI will penetrate deeper into all phases of the software development lifecycle, although the degree of this integration will depend on the maturity of the organizations themselves. Current trends indicate that AI is turning from a disparate “useful fork” into an integral part of a developer’s tools – just as version control systems, or CI/CD, once became the standard.
AI is already involved in many stages: planning and design (AI requirements generation, design creation, architecture sketch)coding (autocomplement, code generation by text description(agency regime))testing (Autogenerate tests, search for bugs)Deployment and escort (AIOps – Incident forecasting, autoscale, log analysis). While these applications are often point – each team itself screws up individual services. But in the future, we can expect a denser end-to-end integration of AI into all SDLC processes. DORA predicts that AI will transform every stage of the lifecycle, from design to support. This means that integrated solutions will appear: for example, your Task Tracker prompts you to clarify the requirement with the help of AI; the development environment automatically generates not only code, but also tests. (Or maybe the test bench lifts.)AI SRE engineer monitors the system and fixes the incident on the sale, and then writes postmortem with the analysis of the causes of the accident.
An important factor is the development of AI models themselves. If 2023–2025 We’ve seen the leap from “slightly helpful” models to solutions that can already solve.70% of standard coding tasks, then 2028–2030 We can see even more intelligent assistants. Most likely, there will be many specialized models: separate for generating UI, separate for optimizing databases, for migrating legashi code, etc. Such narrowly sharpened AI, working as part of a single ecosystem, will make the presence of AI invisible, but permanent at all stages.
What about startups? Young companies have their own dynamics: they are more likely to use AI at all sites where they can to compensate for small staff. It is easier to integrate the latest technologies without bureaucracy. There are already examples of startups where much of the code is generated by AI, testing is fully automated by AI platforms, and people focus on unique logic.
#AI #Engineering #Architecture #ML #Software #Economics #Software