The current trend towards deploying AI agents in companies often goes hand in hand with announcements of major job cuts, as recent reports on Block Inc. and Salesforce show. While technology companies justify these moves as optimisation through AI-supported processes, scepticism is growing over whether they really reflect technological maturity or are instead primarily aimed at increasing shareholder value. Whether AI agents can form part of a sustainable long-term business strategy is difficult to assess at present, as official statements generally remain superficial and offer little insight into how the underlying implementations actually work.
A closer look at Block Inc. illustrates the scale of Jack Dorsey’s ambition. In February 2026, he cut almost half of the company’s workforce with the aim of transforming Block into a flatter organisation built around its own “intelligence layer”. The approach is intended to have a particular impact on software development, replacing conventional structures with company-specific data and capabilities. Salesforce is also relying on proprietary datasets to improve the accuracy of AI outputs to as much as 90%. Yet despite initial cost reductions of 17% in customer service, reports of “reliability problems” are increasing. The AI often fails to meet expectations, suggesting that human involvement remains indispensable in many complex areas.
The main problem lies in the probabilistic nature of LLMs: in critical areas such as programming or research, an element of “chance” introduces risk. According to the Harvard Business Review, AI’s unreliability often creates additional work instead of delivering the expected relief. The current shift towards AI agents therefore looks more like a proactive process of trial and error – in some cases already running into problems, as at Salesforce, and in others, such as Block, with the outcome still uncertain. The production of this newsletter illustrates the same tension: AI can speed up the process, but final fact-checking still has to be carried out manually to ensure reliable quality.
By Marcel Ohrenschall
