Nigel Simpson

Stop Saying “95% of AI Projects Fail” (and Why the viral stat is misleading—and what it really tells us about AI adoption) “Despite $30–40 billion in enterprise investment into GenAI, this report uncovers a surprising result: 95% of organizations are getting zero return.” — MIT/NANDA GenAI Divide, 2025 That line went viral. It’s dramatic, quotable, and fits neatly into a simple narrative: AI is failing. But as always, the truth is more complicated—and far more interesting. After hearing that statistic repeated in every conference and client meeting, I decided to apply a bit of scientific inquiry and read the report myself. What I found wasn’t a story of technological collapse, but one of expectations set impossibly high. report here https://lnkd.in/gyfP4Gwk The authors define “success” as an AI project that reaches production and delivers measurable P&L impact. Hmmm... that’s not a low bar—it’s an Olympic bar! For a technology still in version 1.0 of enterprise adoption, that’s like judging early e-commerce or mobile apps purely by whether they transformed global profit margins in their first year. If we had done that, we might have concluded that 95 % of the cloud failed too. The reality: AI isn’t failing—it’s maturing. And the report itself supports that view. Buried deeper in the findings is a clue to what’s really working: “The most effective AI-buying organizations no longer wait for perfect use cases or central approval. Instead, they drive adoption through distributed experimentation, vendor partnerships, and clear accountability.” That’s the real insight—success favors experimentation, not perfection. In my workshops and client engagements, I’m constantly urging leaders and teams to keep using every tool they can get their hands on—learning daily what works and what can improve. Organizations that learn quickly, partner strategically, and iterate relentlessly are already seeing measurable productivity gains, even if they don’t yet hit the P&L line. And those small, invisible wins? They compound. Every analyst who automates a task, every employee who drafts faster or thinks smarter—each adds a tiny efficiency gain. Over time, those micro-gains add up to massive enterprise value, even if the lines aren’t yet visible on the balance sheet. (See the 1% better rule.) The takeaway: the “95% failure” headline doesn’t prove AI is overhyped. It proves the bar for “success” was drawn too narrowly. We’re still early in the game. The teams that keep learning, iterating, and redefining what success looks like—those are the ones that will win when the score really counts.

13

3 Comments

If LLMs are the way of the future, THEN WHY CAN'T THEY DO LONG MULTIPLICATION? Understanding LLMs' limits on compositionality is *critical* to building large-scale agentic solutions with them. Dziri and Choi's 2023 paper hypothesizes that LLMs solve complex multi-step problems by breaking them down into linearized pattern matching. IF YOU UNDERSTAND THIS LIMITATION, then you can design and engineer solutions around it. BUT YOU MUST UNDERSTAND THIS LIMITATION FIRST. If you imagine that LLMs are "magic" that will just keep getting better and better all by themselves, then you are in for a rude awakening! https://lnkd.in/gxgi3pkD

5

4 Comments

🤖 Ray Kurzweil's humanoid robot startup in talks for $100 million investment ~ Reuters "Beyond Imagination is co-founded by scientist, entrepreneur and filmmaker Harry Kloor. The company has developed a humanoid robot — the Beyond Bot - and accompanying AI models that it intends to deploy in industrial settings such as factories, pharmaceutical plants and chip manufacturing facilities, said Gauntlet Ventures co-founder Oliver Carmack." #robotics #robots #physicalAI #AI #enterprise #venturecapital #funding

11

The Reality Defender API is the antigen test of deep fakes: with two lines of code, you can determine if a bad actor is attempting to spoof your human identity infrastructure (e.g., a fake candidate interviewing for a remote developer role on Zoom, voice identification for your bank, etc.). A game changer in ease of use and operations in preserving human identity.

9

1 Comment

Anthropic unveiled Cowork, a research-preview evolution of Claude Code that lets the Claude AI agent access and modify files in a sandbox folder using natural language - not just generate responses. This marks a shift from classic chat to true productivity agents capable of performing real work on your behalf. Cowork can automate tasks like file organization, document creation, and even multi-step workflows directly on your machine but it has also warned about potential dangers like, if the instructions are not clear then Claude can take "potentially destructive actions" and that with prompt injection attacks, there are a range of potential safety concerns. #agenticai #generativeai #anthropic #claudeai 🔗 https://lnkd.in/gVR7ARWN?

6

I've just published a MailChannels skill for OpenClaw (formerly Clawdbot, then briefly Moltbot). If you have a MailChannels Email API account (signup here: https://lnkd.in/g9e6mzc3), add this skill to OpenClaw and you can start automating email workflows using OpenClaw. And no, we aren't afraid that spammers will sign up. New sources of email abuse emerge thousands of times a day across the hundreds of web hosting providers who use us to send email from their hosting infrastructure. MailChannels is the only email delivery service architected to safely deliver email from uncapped agents like OpenClaw.

19

2 Comments

Want to bring AI into projects but not sure where to start? Manning Publications Co.'s Build AI‑Enhanced Web Apps (https://mng.bz/VVpX) is a practical, hands-on guide that shows exactly how to do it. Instead of focusing on abstract theory, @Theofanis Despoudis walks through real code, UI patterns, and architectural decisions, turning large language models and other AI tools into reliable, production-ready features in modern web applications. It explains how to connect to APIs such as OpenAI and Google, build secure backend services that safely proxy AI requests, design responsive and interactive interfaces that leverage AI, handle streaming responses and real-time feedback, enforce structured outputs with validation and consistency, and add guardrails for safety, bias mitigation, and confidence monitoring, from development through deployment and scaling. Whether the goal is to explore AI as a frontend developer or to ship production features as a full-stack engineer, this book provides a clear, practical roadmap with concise explanations and concrete examples, building skills and confidence to enhance applications with generative intelligence.

7

1 Comment