Essay: AI mania is 'eviscerating' corporate decision-making

What the essay says

An essay published on 18 July 2026 on the blog Ludicity argues that large organisations have lost the ability to make sober decisions about artificial intelligence. Its author, who runs sales and the technical side of client engagements at a consultancy, says the piece draws on roughly 300 conversations with professionals, from small service businesses to Fortune 500 executives. It opens with a warning from Mitchell Hashimoto, known for HashiCorp and Ghostty, that entire companies are under “heavy AI psychosis” and cannot be reasoned with about it.

A 0% success rate, by the author’s count

The central claim is blunt: every AI project the author’s team observed over a year and a half failed — a 0% success rate, including projects the firm was not involved in. The most common failure modes are internal chatbots with almost no uptake, since employees rarely use tools built on poorly documented internal knowledge, and customer-facing bots. The author describes a Mitsubishi voice assistant that promised a callback after a car breakdown; six months later, the call had still not come.

He argues the gap between public claims and internal reality is structural. Employees who speak honestly risk being fired or picked for layoffs, while boards, executives, vendors and consultants all gain from overstating success rates. Some listed companies, he writes, have announced “AI productivity gains” after doing little more than buying Copilot licences.

Loyalty tests and AI-washing

A second section describes how, at companies with more than 500 employees, career progress increasingly depends on public declarations of faith in AI. The essay cites an executive who produced an AI-centred strategy for a business with more than $2 billion in revenue and had never used an AI tool. It also describes “AI-washing”: engineers doing their work as before and attributing it to a model, and teams measured by “token leaderboards” where higher AI spending counts as better.

The author also recounts a sales episode involving Snowflake’s Cortex chatbot. Snowflake staff had put its accuracy at roughly 92% in an ideal configuration — acceptable for a demo, he notes, but not for a finance chief who needs every number to be right. Clients who saw the demo wanted to buy immediately; the team declined the sale and stopped offering it.

Why it matters

The essay traces the situation less to AI itself than to incentives: executives fear contradicting customers’ claims and losing contracts, board members fear looking sceptical, and staff are judged on metrics that are easy to game. It closes with advice for people inside such organisations — prefer one-on-one conversations, use anonymous polls, involve the people who actually use the tools, and avoid challenging the broadest claims about AI in public.