Tim Harford, writing in the Financial Times, recently retold the story of the Vasa — the pride of the 17th-century Swedish navy that sank on its maiden voyage in 1628. The ship sailed for roughly a mile before a gust of wind caught it from the side, water flooded through the open gunports, and 50 metres of ornately carved warship disappeared beneath Stockholm harbour. About 40 people died. The captain was immediately arrested.

Harford draws three lessons from the disaster. The ship was a dangerously innovative design pushed through by a king who was away on campaign. The builders used two different measuring systems — the Amsterdam foot and the Swedish foot — introducing asymmetry that made the vessel lean to port. And the people who knew the ship was unstable chose to sail anyway rather than deliver bad news to the king.

All three lessons have direct parallels in how companies are handling artificial intelligence right now. But the one that matters most is the gap between two groups of people who rarely talk to each other: the staff who use AI every day, and the managers who make decisions about it.

King Gustavus Adolphus wanted an unprecedented number of heavy guns on Vasa. His shipwright, Henrik Hybertsson, knew this would require a second gun deck — something he had never attempted. In an effort to make it work, Hybertsson over-reinforced the upper decks with massive timbers. The result was a ship loaded with heavy guns on a high deck. It was top-heavy from the day the keel was laid.

The parallel is precise. Senior leaders read analyst reports about AI transformation, attend board presentations from consultants, and mandate adoption targets that filter down through the organization. They specify what they want — "deploy AI across customer service by Q3" — without understanding the structural implications of that demand.

Meanwhile, the people actually building with AI — the staff using tools daily — understand exactly where the weight sits. They know which tasks AI handles well and which it fumbles. They know that a chatbot trained on last year's product data gives confident but wrong answers about this year's pricing. They know the difference between a tool that saves two hours a day and one that creates three hours of cleanup work.

These two groups might as well be using different rulers.

Harford highlights a detail that most accounts of the Vasa skip entirely. Four carpenters' rules were found in the wreck, all slightly different lengths. Dutch shipwrights working in a Swedish yard meant the Amsterdam foot (283mm) and the Swedish foot (297mm) were both in use. The resulting asymmetry gave the ship its fatal lean.

Companies measuring AI performance are doing exactly the same thing. Managers track success in deployment metrics — how many teams adopted the tool, how many licenses were activated, what percentage of processes have been "AI-enabled." Staff track success in output quality — did the tool produce work I could actually use, or did I spend 40 minutes editing what it gave me?

The Gartner research that landed this week found 40% of agentic AI projects will be abandoned by 2027. Not because the technology failed, but because organizations deployed it without clear metrics for what success looked like from the user's perspective. Management measured adoption. Staff measured usefulness. Two different rulers. Same ship.

The most tragic detail in the Vasa story is the stability test. A few days before departure, Captain Söfring Hansson arranged a standard demonstration: 30 sailors ran back and forth across the top deck to test the ship's roll. The ship lurched so violently that Vice-Admiral Klas Fleming stopped the test after just three passes. The sailors would normally run more than a dozen times.

Fleming saw the evidence. He had the data. The ship was unstable and everyone on that deck knew it. But the king was away, sending letters demanding Vasa sail immediately. Fleming chose to hope for the best.

This is the dynamic playing out in organizations across every sector. Staff who use AI tools every day run the equivalent of that stability test. They see what works, what breaks, what leans dangerously. But the decision about whether to continue, expand, or cancel the AI program sits with someone who stopped watching after three passes.

The disconnect runs in both directions. Managers kill AI projects that staff find genuinely useful because the ROI metrics don't show up in the quarterly review. And managers mandate AI tools that staff know are worse than the manual process, because the C-suite committed to a vendor contract at a conference in Davos.

The companies getting AI right in 2026 have closed this gap. They have done something that would have saved the Vasa: they let the builders influence the design.

I use Viktor — an AI agent that connects to more than 3,200 tools and does actual work across my publishing operation. Content research, newsletter production, data analysis, document creation. It handles tasks that used to take hours. The reason it works is not that some manager chose it from a vendor shortlist. It works because the person using it — me — selected it based on what I needed it to do, tested it against real workflows, and expanded its role only when the results justified it.

That is the model. Not top-down mandates from people who have never touched the tools. Not bottom-up guerrilla adoption that bypasses security and governance. A deliberate conversation between the people who understand the strategic direction and the people who understand what happens when you actually press the buttons.

You get $100 of free credits to begin. No time limit, no commitment. That's enough to do real work and see what Viktor can actually do before you spend a penny. There's also $50 off your first bill. You must use this exact link to receive both benefits.

The Vasa did not sink because the customer failed to specify that ships should float, as one management paper solemnly suggests. It sank because the person with the authority to stop the launch chose not to act on what the test clearly showed.

Four hundred years later, the choice is the same. Listen to the people doing the building, or hope for the best and watch from the shore.

Disclosure: Some links in this article are affiliate links. If you choose to get started with Viktor using the links provided, I may receive a commission — at no additional cost to you. I only recommend tools I use and believe in.