# Why AI rollouts die in the IT department - and 4 plays that actually work | Open Mercato Source: https://www.openmercato.com/blog/why-ai-rollouts-die-in-the-it-department # Why AI rollouts die in the IT department - and 4 plays that actually work The board catches FOMO, then the energy slams into the IT wall. That pushback is rational, not obstruction. Here are four plays we keep seeing in teams that roll out AI well — starting with a list of everything that could go wrong. - Tomasz Karwatka - June 3, 2026 Software is about to be built completely differently Explore with AI - [Ask Claude about this article (opens in a new tab)](https://claude.ai/new?q=Read%20this%20article%20so%20I%20can%20ask%20questions%20about%20it%3A%20%22Why%20AI%20rollouts%20die%20in%20the%20IT%20department%20-%20and%204%20plays%20that%20actually%20work%22%20(https%3A%2F%2Fwww.openmercato.com%2Fblog%2Fwhy-ai-rollouts-die-in-the-it-department).%20Start%20with%20a%20short%20summary%20of%20what%20it%20says.) - [Ask ChatGPT about this article (opens in a new tab)](https://chatgpt.com/?q=Read%20this%20article%20so%20I%20can%20ask%20questions%20about%20it%3A%20%22Why%20AI%20rollouts%20die%20in%20the%20IT%20department%20-%20and%204%20plays%20that%20actually%20work%22%20(https%3A%2F%2Fwww.openmercato.com%2Fblog%2Fwhy-ai-rollouts-die-in-the-it-department).%20Start%20with%20a%20short%20summary%20of%20what%20it%20says.) - [Ask Perplexity about this article (opens in a new tab)](https://www.perplexity.ai/search/?q=Read%20this%20article%20so%20I%20can%20ask%20questions%20about%20it%3A%20%22Why%20AI%20rollouts%20die%20in%20the%20IT%20department%20-%20and%204%20plays%20that%20actually%20work%22%20(https%3A%2F%2Fwww.openmercato.com%2Fblog%2Fwhy-ai-rollouts-die-in-the-it-department).%20Start%20with%20a%20short%20summary%20of%20what%20it%20says.) In most mid-sized and large companies, learning to use AI follows the same script. The board catches a healthy case of FOMO. The CEO walks in and asks: “Folks, is anything passing us by here?” There’s energy. There’s appetite to move. And then that energy hits a wall - the IT department. Here’s the thing: the people who actually do the work know it better than their bosses. In five minutes they can produce a wave of arguments for why this won’t work. That’s not irrational resistance. It’s a rational concern about the quality of the work. IT knows the pattern: everyone is in a hurry today, but when it breaks, they’re the ones left holding the problem. At Open Mercato we keep meeting companies that handle AI adoption better than the rest - especially at the level of whole teams. Here’s what they do differently. ## 1. Start with the list of things that could go wrong Counterintuitive, we know. But when we sit down with IT and start writing out what might break - ourselves, first - we reach alignment faster. People say: “finally, someone reasonable.” You move out of “who’s right” mode and into “shared exploration” mode. The list isn’t a blocker. It’s the on-ramp. ## 2. Don’t transform all of IT at once. Build a pirate ship. The best rollouts we’ve seen all look the same: a small team running alongside the main org - maybe 10–20% the size of the “tanker.” Often reporting straight to the CEO. Volunteers only. The commandos who _wanted_ to be there. Classic IT stays the source of truth - your SAP, your data. The pirate ship builds new AI applications on top of it. It works surprisingly well, precisely because nobody is asking the tanker to turn on a dime. ## 3. POC mirroring. Nothing ships to production. Our favorite trick. You build a bypass: take, say, 10% of the work - the same customer ticket - and route it in parallel to an AI agent _and_ to a human. The human does their thing. The AI does its thing. Nothing goes live. Then the human compares the output. You collect data for a month. And with that data, IT walks into the board for a concrete conversation - not a vibes-based one. What usually happens: AI handles around 30% of cases. And there’s a subset of tickets, customers, or case types it solves brilliantly. That’s where you keep digging. The rest, you leave alone. ## 4. This is a game played on the relationship, not the license We’ll say this even against our own short-term interest. In theory we should just sell you a license. But if we sell the license and the project falls apart a month later - you won’t renew it. So: slower, on data, on trust. Long term, this is about the relationship. A few years ago nobody bet that IT would be the first function under pressure from AI. Everyone said “we’ll be safe.” That’s probably part of why there’s so much emotion in these teams today. Written by Tomasz Karwatka CEO and Co-Founder ## More from the blog ### [Cezar: how we got back control over our AI coding agents](https://www.openmercato.com/blog/cezar-cockpit-for-ai-coding-agents) A deep dive into Cezar, our open source cockpit for AI coding agents: queue 30 tasks, set one limit, close the laptop. Architecture, Actions, autofix workflows and three ways to run it. ### [We benchmarked coding agents on Open Mercato. The hardest failure looked perfect.](https://www.openmercato.com/blog/coding-agents-benchmark-open-mercato) Three open-weight models built real Open Mercato features. The first benchmark run showed where agents are already reliable, and where convention-perfect code still breaks at runtime. ### [Eleven Questions CTOs Ask Before They Build on Open Mercato](https://www.openmercato.com/blog/cto-questions-before-building-on-open-mercato) The questions that come up in almost every technical due diligence call: versioning, traceability, upgrades, contributor QA, token costs, scale and the business model behind an MIT license. With the answers we give in the room, including the ones where the honest reply is that we have not done it yet. ## Software is about to be built completely differently.