Generative AI is everywhere and it’s tempting to reach for it whenever something feels messy, slow, or frustrating.
But when a tool is this powerful – and this non-deterministic – the real question isn’t “Can we use GenAI?” It’s “Should we?”
Used well, GenAI boosts productivity. Used indiscriminately, it quietly introduces risk.
This is where GenAI stops being just a productivity tool and starts becoming a governance challenge.
After more than a year, on and off, building agents across LangFlow, Microsoft Agent Framework, and Copilot Studio – from PoCs to my own real-world deployments – one theme keeps nagging at me: prompt debugging feels like a black box adventure.
In traditional software development, you can step through the code, trace errors, and monitor state changes with powerful tools. But with natural language programming? You’re trusting your instructions to a probabilistic model whose reasoning you rarely get to see.
And that changes everything.
In the ever changing world of enterprise GenAI, the new Researcher Agent functionality in Microsoft 365 Copilot started me questioning whether I should retire my own Copilot Studio developed M365 Research Agent. So, I tested it and really only found one minor flaw (that I couldn’t select sub-folders from SharePoint sites).
I’ve been experimenting with Microsoft’s new Agent Framework (MAF) – but instead of connecting to cloud systems, I’ve been running it entirely offline on an Amazon EC2, private cloud, instance. My goal was to see whether this new, unified framework could function offline, be used with offline LLM’s and process PDFs (of RFPs in this case), extract questions, and even draft answers – all without leaving a secure, private environment.
It worked remarkably well. But what’s even more interesting is what this means for organizations on multiple fronts: the ability to run sophisticated Agent workflows locally, maintain full control of data, and start automating complex knowledge tasks such as RFP responses, compliance checks, or policy reviews.
Back in 2024 (it seems so long ago now) I wrote about Agents (links below) and cautioned about how early we were in their evolution. Now, almost a year later we seem to be in a completely different place – brought back to my mind to revisit by the recent announcements from:
– Langflow – releasing v1.6
– Microsoft – consolidating AutoGen and Semantic Kernel into the Microsoft Agent Framework
– OpenAI – releasing AgentKit
Big AI models often steal the spotlight, but sometimes the smartest move is going smaller. Google’s new Gemma3 270M shows just how powerful a compact, efficient language model can be – especially when it runs offline, on low-power devices, or in remote locations. For businesses, this isn’t just a technical breakthrough; it’s a new frontier of opportunity.
Microsoft’s recent Build 2025 announcements have brought some massive updates to the whole GenAI Copilot platform. One of the most exciting features for me (and perhaps a revealing feature in terms of Microsoft’s GenAI strategy) is the introduction of Microsoft 365 Copilot Tuning. This new feature is set to revolutionise how organisations tailor AI to their specific needs, moving powerful model fine-tuning capabilities from the often complex of data scientists and technical staff to the more accessible world of business power users.
