{"id":17538,"date":"2025-02-28T13:24:25","date_gmt":"2025-02-28T12:24:25","guid":{"rendered":"https:\/\/www.centigrade.de\/?post_type=blog&p=17538"},"modified":"2025-03-05T10:08:24","modified_gmt":"2025-03-05T09:08:24","slug":"human-centred-product-development-with-ai-agents-or-the-industrialisation-of-digitalisation","status":"publish","type":"blog","link":"https:\/\/www-test.centigrade.de\/en\/blog\/human-centred-product-development-with-ai-agents-or-the-industrialisation-of-digitalisation\/","title":{"rendered":"Human-centred product development with AI agents or … the industrialisation of digitalisation"},"content":{"rendered":"

\"KI<\/p>\n

The arrival of generative AI and the widespread use of ML algorithms is already significantly changing our human behaviour and organisational processes at all levels of society – and at breakneck speed.<\/strong><\/p>\n

<\/p>\n

The way in which we will develop digital products in the future is changing just as radically. As large language models (LLMs) are very familiar with syntax and grammar, the generation of computer source code is already common practice and the productivity of companies that use AI in this area is experiencing a measurable increase<\/a> compared to those that are still working \u2018by hand\u2019.<\/p>\n

But if source code is no longer written by humans, but by AI agents, is it still worth learning the profession of software engineer? Nvidia CEO Jensen Huang thinks not<\/a>:<\/p>\n

\n

“We will make computers more intelligent so that people don’t have to learn computer science to programme a computer.”<\/em><\/p>\n<\/blockquote>\n

Even Mark Zuckerberg, who is known for his technical realism, recently proclaimed the end of the human mid-level engineer<\/a>. But if even the profession of software engineer is being jeopardised by generative AI, what does this mean for the digital product development of the future in general? Shouldn’t all processes, frameworks and methods that deal with the design and realisation of software also be called into question?<\/p>\n

After all, current approaches could have fundamental disadvantages in terms of speed and efficiency because they were essentially developed for human collaboration and not for \u2018agentic\u2019 collaboration. So how will human-centred digital product development change at the next evolutionary stage of generative AI – the stage of autonomous AI agents?<\/p>\n

AI agents – the next evolutionary stage of generative AI<\/h2>\n

A new actor enters the stage<\/h3>\n

On social media, the year of AI agents has already been intensively heralded as 2025. It is a kind of \u2018next level\u2019 within the still relatively young AI hype cycle<\/a>. Instead of humans writing prompts to realise their intentions, AI agents act largely autonomously. They plan, execute, reflect and improve. Some AI agents come to their own conclusions and others act in secret – as \u2018secret agents\u2019, so to speak.<\/p>\n

\"cartoon

AI agents also sometimes work in secret and don’t always do what they’re supposed to. Source: Continuous UX Fun<\/a><\/p><\/div>\n

Obviously, it makes a huge difference whether a software engineer – when he is stuck again – enters his problem as a prompt in ChatGPT, has the corresponding source code generated there and then integrates it into his code in parts or as a whole, or whether an AI agent generates source code autonomously and commits it independently, as if it were a \u2018real\u2019 software engineer.<\/p>\n

The main difference between classic \u2018prompt-based\u2019 generative AI and so-called \u2018agentic\u2019 generative AI is, among other things, whether the initiating actor is a human or an AI and to what extent a human or an AI produces and processes the result. Since AI agents have a memory, pursue goals, plan, reflect and respond, they can in principle become active at any point in an information processing chain and not just in the middle, as the following table illustrates.<\/p>\n

\n\n\n\n\t\n\n\t\n\t\n\t\n\t
Evolutionary stage<\/th>Initiator<\/th>Producer<\/th>Further processors<\/th>\n<\/tr>\n<\/thead>\n
Without AI<\/td>Human<\/td>Human<\/td>Human<\/td>\n<\/tr>\n
\u201ePrompted AI\u201c<\/td>Human<\/td>AI<\/td>Human<\/td>\n<\/tr>\n
\u201eAssisting AI\u201c<\/td>AI<\/td>AI<\/td>Human<\/td>\n<\/tr>\n
\u201eAgentic AI\u201c<\/td>AI<\/td>AI<\/td>AI<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/p>\n

Strong together: multi-agents in a team<\/h3>\n

But AI agents can do far more than just produce source code: They can communicate with each other, operate tools and make joint decisions, which ultimately enables them to form teams and swarm intelligences. The \u2018multi-agent\u2019 framework MetaGPT<\/a>, for example, initiated by DeepWisdom.ai, has implemented an entire product development team of AI agents and allows this AI team to jointly develop software. This approach makes it clear once again that the development of software products cannot be limited to writing source code. In addition to the software engineer, a team of AI agents also includes a product manager, a software architect, a project manager and a quality assurance manager. There is even the obligatory \u2018boss\u2019.<\/p>\n