If you’ve been using GenAI tools like Microsoft Copilot or ChatGPT in your day-to-day work, you’ve probably had this experience: something that used to work, like a prompt you carefully refined, is suddenly behaving differently. Maybe it’s not as helpful. Maybe it’s giving unexpected results (that’s what happened to me this week). Maybe it just… stopped working entirely.
This isn’t just annoying, it’s revealing something deeper about the nature of GenAI adoption that we don’t talk about enough and what it means to how you approach work with GenAI.
The changes in this field are fast, but the change is also wide and deep. It’s reshaping work across every department, and just as importantly, it’s continuing to change things we already thought were “done.” That’s where the real disruption lies, not just in new tools, but in the quiet, unpredictable shifts in the ones we’re already using.
What Is It?
With most software, once something works, it stays that way – we rely on that consistency when we are trying to get work done. Systems are versioned, changes are scheduled, and users can build processes around stable functions – GenAI isn’t working out quite like that.
Under the hood, these systems are powered by massive language models and functionality wrapped around them that are frequently updated – sometimes by vendors, sometimes automatically through cloud deployments. A prompt that worked last week might respond differently today, not because you changed it, but because the model changed. Often, there’s not sufficient notification, no changelog, and no way to roll back.
Even small tweaks to how a model interprets tone, structure, or context can quietly break user workflows. And because GenAI is inherently non-deterministic, outputs can vary even when nothing changes.
This means we’re dealing with tools that are not just powerful, but volatile.
What This Means from a Business Perspective
This volatility introduces a real problem for business operations: inconsistency.
When GenAI tools are embedded into business processes, whether that’s summarising emails (my use-case that prompted me to put this article together), generating reports, or drafting internal documents – consistency matters. People rely on those processes to save time, reduce risk, and deliver reliable outputs. If the same prompt gives different results week to week, confidence erodes. Teams hesitate. Workflows fall back to manual. The expected efficiency gains stall.
And this happens across the business. Marketing, Legal, Customer Service and IT. Everyone is affected. The breadth and depth of GenAI’s reach makes these inconsistencies more than isolated incidents, they become systemic.
The result? A sense of instability that makes it harder to scale adoption, enforce governance, or confidently build long-term value around GenAI.
What do I do with it?
So what’s the answer? It’s not to slow down adoption, but to adapt how we adopt.
Here are a few concrete steps organisations can take:
- Expect change, design for resilience. Treat GenAI tools as dynamic systems, not static software. Build in checkpoints for regular prompt reviews and validations.
- Create feedback loops. Encourage teams to document and share when things break or behave differently. Internal Communities of Practice (CoP’s) are powerful mechanisms to crowdsource solutions and keep people aligned.
- Focus on principles, not prescriptions. Train users to think in terms of outcomes and intent, not just specific prompts. A flexible mindset helps teams recover faster when tools shift.
- Push for transparency. Work with vendors to understand model changes and push for clearer update logs, changelogs, or sandbox environments for testing. For example Microsoft has the 365 Roadmap where you can research upcoming changes.
- Embed governance lightly but early. Track how GenAI is used, ensure proper controls are in place, and make sure someone owns the responsibility for reviewing impact as tools evolve.
We often talk about GenAI being fast, but the real challenge is that it’s fast, broad, and deep. That means the ground is always shifting – sometimes subtly, sometimes dramatically. If we want to build durable value from these tools, we need to accept that solid ground isn’t under us at the moment and we need to learn how to move with it.
Further Reading
The Outlook and Copilot Video that clarified things for me (Thanks Nicholas Harris)
Copilot Mechanics Blog (from Microsoft)