Use of the term ‘we’re implementing AI’ is now ubiquitous across much of the corporate world. Increasingly, it is also becoming meaningless.
Andy Jassy, president and CEO of Amazon, told staff in a memo in June 2025 that the company had more than 1,000 generative-AI projects under way and that the technology would shrink its corporate workforce within a few years. Arvind Krishna had put it more plainly six weeks earlier, telling the Wall Street Journal that IBM’s AskHR agent had taken over the work of a few hundred personnel staff, even as the firm’s total headcount rose because the savings went on programmers and salespeople. By the autumn the talk had a measure. John Butters, an analyst at FactSet, counted the phrase ‘AI’ on a record 331 earnings calls between September and December, six times the level before ChatGPT arrived. The firms that mentioned it gained less in the market than those that said nothing.
Tobi Lütke tried to give the expectation teeth. In a memo in March 2025, Shopify’s chief executive told staff that using AI well had become “a fundamental expectation”, and that managers must prove a job could not be done by software before hiring a person. Yolanda Seals-Coffield had been set a gentler version of the brief at PwC, where the chief people officer was to train 75,000 American employees on the tools, part of a billion-dollar plan the firm announced in 2023. Swami Sivasubramanian of Amazon Web Services had gone wider still, promising in December 2023 to train two million people through courses that were mostly self-paced and open to anyone, on the payroll or not.
Aditya Challapally found the gap between the announcing and the doing. In an MIT study published in August 2025, he examined 300 corporate deployments and found that 95 per cent of generative-AI pilots had moved no profit, against spending he put at between 30 billion and 40 billion US dollars.
The fault, he reckoned, lay not in the models but in a “learning gap”: firms had named the technology without weaving it into how they worked, and his team kept finding staff who used ChatGPT off their own backs while the sanctioned pilots stalled. Marc Benioff had been less troubled by any of this. Salesforce’s chief executive told Bloomberg in June 2025 that AI already did between 30 and 50 per cent of the work at the firm, and by September he had cut its support team from 9,000 people to about 5,000, telling a podcast he needed fewer heads.
Word and deed
Sebastian Siemiatkowski had gone furthest of any of them. Klarna’s chief executive had frozen hiring in 2024 and said an OpenAI chatbot was doing the work of 700 agents; by May 2025 he was rehiring people, telling Bloomberg the automated service had been “lower quality” and that customers still wanted a person to talk to. Luis von Ahn had travelled the same road more quickly, calling Duolingo “AI-first” in late April 2025 and then retreating within weeks, once staff and longtime users turned on him.
Gary Gensler had already drawn a line under the looser version of the claim. In March 2024 the Securities and Exchange Commission, which he then chaired, charged two investment advisers, Delphia and Global Predictions, for selling an AI capability they did not have. Gurbir Grewal, his enforcement chief, called the practice “AI washing”; the commission reached its first public company, Presto Automation, the following January. The charges went after false statements rather than the fashionable word, but they meant a later boast could be tested once it reached a filing.
Von Ahn changed course inside the year. His staff at Duolingo had begun asking whether the company wanted AI for its own sake or for something it actually did. In April 2026 he dropped the rule that had tied their performance reviews to how much of it they used.
Photo: Dreamstime.