Happy New Year! It is wonderful to be back with you for the first time in 2026. Looking back, 2025 was a year of profound duality — as satisfying and rewarding as it was demanding and frustrating. As we move into the months ahead, my goal is for this newsletter to serve as your stable anchor.
Whatever 2026 brings, I am certain it won’t be dull and I am glad we are navigating it together.
The Hangover After the Party
The champagne has gone flat. After $30–40 billion poured into AI experiments, most enterprises are waking up with a headache and not much to show for it. MIT researchers call this “The Great Recalibration.” The problem is not that AI models are not smart enough — it is that our organisations are too rigid to absorb them. We have been trying to pour water into a jar that is already full.
This issue breaks down why most companies are stuck in what I call “Pilot Limbo” and what the 5% who escaped actually did differently.
A. Pilot Limbo: Where Good Ideas Go to Die
Picture a gym in January. Packed with new members, everyone full of optimism. By March, the treadmills stand empty, replaced by digital tumbleweeds as initiatives fizzle out. That is what is happening with AI pilots across the enterprise.
The MIT report shows that while 80% of companies have experimented with AI, the drop-off from “playing around” to “actually working” is a steep cliff — as only 5% of pilots ever see the light of production and 95% remain in development hell due to:
1. The Goldfish Problem
Most AI systems have the memory of a goldfish. Every conversation starts from zero. They do not learn your company’s nuances, remember last week’s decisions or build on yesterday’s work. Users end up re-explaining context over and over like training a new intern every single morning. Eventually, people give up and go back to doing things the old way. You can’t blame them entirely — so AI sits unused as an expensive reminder of good intentions.
2. Chasing the Shiny Object
There is also a massive misallocation of resources. Companies are pouring 50–70% of their AI budgets into Sales and Marketing, which seems glamorous and visible — but these are the hardest areas to automate because human relationships are messy and unpredictable.
Meanwhile, the real gold is in the “boring” back office: processing invoices, checking compliance, handling procurement — areas with structured data, repeatable processes, and verifiable outcomes that deliver reliable ROI. The 5% who made it focused here and eliminated millions in outsourcing costs.
B. The Speed Trap: When Faster Feels Slower
A real randomised trial found that experienced developers actually slowed down when using AI coding tools — even though they believed they were faster. AI is brilliant at getting you 70% of the way there, scaffolding code fast and generating boilerplate in seconds, but that last 30% — integration, edge cases, and system compatibility — becomes harder, not easier. What felt like flying often turned into longer review and debugging cycles.
C. Agent Washing: The Emperor’s New Clothes
“Agentic AI” has become the new buzzword. Vendors promise systems that can plan, execute and handle complex tasks autonomously — sounding like science fiction brought to life. But Gartner predicts 40% of these projects will be abandoned by 2027 because most “agents” are wolves in sheep’s clothing — basic tools dressed up in marketing language. Level 4 agents fail nearly 70% of multi-step tasks, where one small error can cascade into larger problems.
D. The Underground Railroad of AI
While IT departments debate enterprise licences and security protocols, 90% of your workforce are using personal AI tools and whatever gets the job done. It is the prohibition of AI where people want the tools — official channels are slow, clunky or nonexistent. So they find their own supply. The result is a massive “Shadow AI Economy” operating outside governance.
This should terrify you:
- Your secrets are leaking: employees paste confidential data into consumer AI tools that may use inputs for training.
- You are flying blind: productivity gains are invisible and the company captures none of the learning or improvements.
- You are one password away from disaster: critical workflows may depend on personal accounts that can be lost or abandoned.
The solution is not to ban personal tools — that approach already failed. Make your official tools better than the bootleg version: pre-load them with company context, make them smarter about your data, and give people a reason to come in from the cold.
E. What to Do About It
Five moves for Q1 2026:
- Kill the zombies: audit every AI pilot with a “Net Lift” formula — Old Way minus AI outcomes, human review, and fixing mistakes.
- Follow the money: redirect budget from flashy experiments to back-office automation.
- Fix the plumbing before buying appliances: the real blocker for AI isn’t model intelligence — it is messy data. Clean your APIs and knowledge graphs so systems are “agent-ready.”
- Demand receipts from vendors: ask which level their tool really operates at, not marketing jargon.
- Build the cleanup crew: create dedicated roles to review AI output and reward quality over volume.
These moves help organisations move out of Pilot Limbo and make AI deployment sustainable, measurable and impactful.
