Practical Insights on
Enterprise GenAI

As I've seen new technologies in the GenAI space being released over the last few years I have pondered the effect on organisations and decided to take a look at the impacts through the lens of two classical strategy models - Michael Porter's 5-Forces and Value Chain.

I am about to make two of my codebases public. One is the multi-agent RFP responder I have posted about before (which I use to respond to RFP's). The other is an RFP assessor. I spent a fair bit of time thinking about whether I should try to turn them into products. There was a period where that felt like the natural next step. Build something useful, prove that it works, then package it up. But the more I reflected on it, the more I found myself landing somewhere else.

A lot of GenAI discussion is still in the “help me do this task faster” phase. That made sense for the first wave. But we are now moving to something bigger and more disruptive. First, GenAI helped people with tasks. Now it is starting to take on processes. Next, it will start to reshape, compress, and in some cases replace parts of, or complete roles. I do not think most organisations are preparing for that shift with anything like the seriousness it deserves.

Last weekend I was down a logprobs rabbit hole, trying to measure model confidence a bit more honestly, this weekend was about visualising it.

One of the most important mindset shifts with Generative AI is that hallucinations are not a strange side effect that appears only when a model “goes bad.” They are tied to how these systems work. Large language models generate text by predicting the next token, not by checking whether a statement is true in the way a rules engine would. That is why they can sometimes produce something fluent, plausible, but wrong. The goal is not to pretend this problem will disappear. The goal is to learn how to work with it responsibly.

I have been a bit uneasy for a while about asking a model to rate its own confidence (at least it's a data point). It can be useful, but it also feels a bit circular. If the same system can hallucinate the answer, it can also hallucinate the confidence score. So this weekend I went down a logprobs rabbit hole.

Some organisations start their Generative AI journey by talking about tools, policy, governance, or risk. Those things matter. But in practice, the biggest driver of adoption is often much simpler: AI literacy.

Today’s experiment turned into a bit more of an odyssey than I expected. On paper, simple enough: • Test a lighter-weight multilingual model as an alternative to Granite4 3B • Run it through the 500 prompt injection attack set to get a baseline • Fine-tune it using the new Unsloth AI Studio tool • Rerun the test and compare the results

As organisations move from AI assistants to more agentic workflows, the risk picture starts to change. A document is no longer just something an employee reads. It may now be something an AI system reads, interprets, and acts on. That sounds efficient, and often it is. But it also creates a new control question for GRC teams: what if the document contains instructions designed to manipulate the workflow itself?

A risk with any system that ingests uploaded documents and passes extracted content through LLM-driven workflows is that the document itself may contain prompt injection attacks. In other words, the RFx is not always just a source of requirements - it could also be a delivery mechanism for malicious instructions aimed at the downstream agents. So this weekend I added a dedicated Security Agent into the pipeline.

Most people still think of AI in organisations as a better assistant: something you prompt, something that helps, something that saves a bit of time. But the real shift underway is much bigger. We are moving from prompting AI to truly delegating work to AI - and the path to digital coworkers is closer than most organisations realise. Microsoft’s new Copilot Cowork is a strong signal of that direction, positioning itself as an execution layer for M365 that can turn intent into action and still keep the user in control.

For a long time, software decisions have been framed as a fairly binary choice: do we build something ourselves, or do we buy it from a vendor? That framing still exists, but needs to be expanded. With the rise of AI coding tools, workflow orchestration, and the possibility of systems that can generate logic at runtime, the choices has become far broader - and far more interesting.

his weekend’s experiment tackled a problem I’ve been curious about for a while. I had a request for an AI-driven, Google-based solution to scan websites and draft emails from the content. I wondered whether an AI vision-based approach could reliably extract data from modern websites where traditional scraping may fail?

We are moving from "Active AI" (where you ask a bot to do something) to "Ambient AI" (where the system anticipates the need and does it for you) - another way of putting this is Autonomous Agents running in the background, becoming ambient.

GenAI (and especially agents) will reduce the time and effort involved in drafting RFx documents and responses. That said, the biggest constraints in procurement usually sit elsewhere - the bottleneck tends to shift rather than disappear. Working on my own RFx response and RFx assessment agents started me thinking, that instead of a sequence of documents and meetings (Requirements → RFx → Response → Evaluation → Negotiation → Contract management → Close), procurement starts to behave more like a pipeline, a continuous workflow where structured outputs flow from one stage to the next, and get reused across cycles.

This weekend’s experiment was conceptually simple: 𝘤𝘢𝘯 𝘐 𝘶𝘴𝘦 “𝘨𝘰𝘰𝘥 𝘥𝘦𝘴𝘪𝘨𝘯” 𝘪𝘯 𝘰𝘯𝘦 𝘤𝘰𝘥𝘦𝘣𝘢𝘴𝘦 𝘵𝘰 𝘳𝘦𝘧𝘢𝘤𝘵𝘰𝘳 𝘢 𝘗𝘰𝘊 𝘪𝘯 𝘢𝘯𝘰𝘵𝘩𝘦𝘳 𝘤𝘰𝘥𝘦𝘣𝘢𝘴𝘦?

A LinkedIn post from Clark University’s advancement team stopped me mid-scroll - not because “7 AI agents” is technically significant, but because it’s a new kind of organizational announcement. They describe software components the way you’d describe hires: clear roles, scopes, budgets, governance, and “human oversight”… plus an explicit boundary around relationship work. It’s a glimpse of how automation can be socialized inside organizations.

As there has been lots of posting around the Anthropic Opus 4.6 model I decided to get around to a full analysis of 𝗺𝘆 𝗥𝗙𝗣 𝗥𝗲𝘀𝗽𝗼𝗻𝗱𝗲𝗿 to review the codebase and identify gaps in a few areas.

Rolling out Generative AI in the workplace is more about people than platforms. Over the past year and half, I’ve helped a number of organisations launch GenAI initiatives - and nearly every one of them has surfaced questions, worries, or resistance from staff (with some common themes). These concerns are not signs of failure; they’re signs that people are paying attention. In this article, I want to share the most common concerns I’ve encountered - and how organisations can respond in ways that build trust, not tension.

This week I re-asserted an architectural decision - and it drove some database work (I went with Postgres).

Agentic coding tools (like Claude Code, OpenAI’s Codex agents) are making it ridiculously easy to turn an idea into working software. That’s exciting. It’s also where people can get into trouble - especially when non-developers or non-solution designers use these tools to build systems they can’t confidently secure, test, operate, or maintain. Below is a pragmatic way to think about agentic tools: when they’re a superpower, when they’re a liability, and how to get value without accidentally creating a future incident (or an unmaintainable mess).

This week I got a clean end-to-end run of my RFP Factory: upload through the Web UI, multi-agent orchestration does the heavy lifting, and a complete draft package comes out the other side. It’s one thing to have individual components working; it feels like a whole other step to have the whole pipeline behave like a real system.

If you follow the GenAI conversation closely, it can feel like whiplash - like there is no agreement. One day it’s “AI is rewriting the economy,” the next it’s “AI is all hype and risk.” It feels like we’re now in a “Great Divergence” - not just differing opinions, but two parallel realities shaped by incentives and where you sit in the organization (something I have seen in my own work - some organisations are embracing AI and realizing the benefits while other are flatly, not interested).

Today was a two-track build day: getting deeper into OpenRAG, and putting a simple UI in front of my agent stack.

The newest wave of “desktop automation” tools look genuinely useful - and materially different from the assistants we’ve gotten used to. Tools like Claude Cowork and agentic browsers such as Perplexity Comet and ChatGPT Atlas don’t just answer questions; they can take actions across your files, tabs, and workflows. That shift changes the risk profile, fast.

This week focused on performance benchmarking, architectural consolidation, and refining the quality of the automated response across the 5 agents (soon to be 6). All still inside Microsoft Agent Framework (MAF) with A2A and MCP.

Most of my recent Coffee & Code time has been going into one very specific thing, improving the RAG layer behind my Microsoft Agent Framework–based RFP multi-agent solution.

GenAI is an accelerant. It speeds up decisions, output creation, and information flow, often without strengthening the system underneath. And many organisations are already running “hot”: highly optimised, tightly interconnected, little slack, and dependent on tacit knowledge. So the real question isn’t just “How much can we automate?” It’s also “Where does speed strengthen the system - and where does speed increase fragility?”

Responding to RFPs used to feel like running a marathon (it's just as painful as being on the RFP assessment team) - days of effort, multiple people, and thousands in costs. Recently, I asked myself: Could AI make this easier? What started as an experiment (we are always experimenting with the edge of this technology) with Microsoft’s Agent Framework on a local setup evolved into a multi-agent orchestration system that drafts RFP responses in under 15 minutes.

Less time available this week than usual but still have progress.

Today's job was to split the existing PoC into a multi-agent system, making the RFP Responder system easier to manage, understand conceptually, architecturally sound and learn about A2A with Microsoft MAF.

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.

Last week, I shared the first working version of my offline PoC RFP Agent Factory - using Microsoft Agent Framework, Ollama, ChromaDB and Docling MCP (all the real focus of the work) to autonomously and automatically generate draft RFP Responses and Response Assessments.

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.

One of the challenges about GenAI adoption is simply getting started: picking tools, running pilots, training staff, and rolling out a plan. Another major challenge is where and how GenAI gets introduced into already fragile, tightly coupled organisational systems. I was watching a Veritasium video (The Strange Math That Predicts (Almost) Anything) about complex systems and the moment they reach a “critical state.” A forest can look calm and stable right up until a single spark turns it into a massive wildfire. Not because the spark was special but because the system was already primed for runaway behaviour. In a general sense, many organisations today look just like that forest, in a critical state.

A little while back, I posted about an experiment: running the Microsoft Agent Framework (MAF) entirely offline on an Amazon Web Services (AWS) EC2 instance, paired with Ollama and a Docling MCP server. The goal was simple - test whether MAF could operate in a fully contained environment with no external API calls. It worked.

Most of the time, when we talk about large language models (LLMs), we end up in the weeds of training data and parameter counts. Useful if you’re a researcher; less useful if you’re a leader, policymaker, or practitioner trying to answer a simpler question: “Is this thing actually behaving in a way I’m comfortable with?” Two realities make that hard: The training data is too large for humans to grasp in any meaningful way. The models are too complex for us to truly understand their internal “decision making.” But their outputs - the words they put on the page - are something we can read, interrogate, and assess.

All organisations seem to be dealing with the same question - “Where do we even start with GenAI?” But the context behind that question is very different. In large organisations, there are budgets, teams, governance committees, structured programs and projects. In small businesses, there’s you, a small team, and the pressure of everyday operations. This article looks at why GenAI adoption isn’t just a scaled-down version of enterprise AI adoption - and why small businesses need a different, more streamlined approach.

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.

Last week I explored the Boston Crime Statistics dataset (~260,000 rows) using Excel Agent Mode, which lives within Excel. This week I revisited the same dataset with the same question using Microsoft’s M365 Analyst Agent - it is a completely different experience. Both tools analyze data and generate insights, but they differ in how you interact with them and how they talk, and show their work. One keeps you grounded in the familiar grid of Excel; the other lifts you into a conversational workspace (real Conversational Data Analytics) that feels more like working with a colleague than a formula bar.

I’ve been experimenting with the Microsoft's new MAF - but with a twist. Instead of using cloud models or online APIs, I set out to see whether MAF could operate completely offline using local components only.

Earlier this year I compared Google Colab and Excel Copilot for analyzing Boston Crime Statistics (Google Colab vs Excel Copilot). This time I tried the same data set with Excel Labs Agent Mode and it was a completely different experience in Excel. With the same dataset - 260,000 records of Boston crime incidents - and the difference is night and day. Where Copilot stumbled and failed, while Agent Mode delivered a complete analysis with explanations and recommendations, all while staying comfortably within Excel.

It used to be great to find others that 'spoke Excel' - understood the intricacies of the various lookup formulas or when to use index...match. I have spent some time working with Excel Labs Agent Mode and the 'old' Excel world is about to change dramatically. Excel Agent Mode has arrived as part of Microsoft's Frontier preview program, and after testing it myself to create survey data for my training courses, I can see this isn't just another incremental update (spreadsheet link included in Further Reading section below). It might make those ribbon menus obsolete.

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

I've spent the past few months experimenting and researching here and there with tiny and small language models, e.g. running log analysis on edge devices, processing audio in remote locations where connectivity is spotty, power is low and the environment harsh. They're fast, efficient, and honestly? Pretty fun to work with and research. But lately, I've caught myself asking: Am I actually solving a problem here - or just doing something because it's technically interesting? If you're working with AI in any capacity, you've probably felt this tension too (and to be honest, sometimes because something is technically interesting, that can be a good enough reason for personal research).

Three critical components of the AI ecosystem are accelerating in different directions with no gravity to bind them. Technology development races ahead exponentially, legislation struggles to keep pace, and business adoption moves cautiously behind them, creating an unstable system that threatens our competitive positions.

Your team just ran a successful GenAI pilot – but how do you keep the momentum going? Too often, early wins fade without a way to spread knowledge and enthusiasm. A Community of Practice (CoP) is one of the best ways to make GenAI adoption take root and grow.

Imagine giving two employees the same task, but one has all the right tools and the other doesn’t. You’d get very different results. I experienced the same with Microsoft’s Copilot Studio agents – and learned that how you build your AI assistant makes a huge difference in what answers you get from SharePoint.

Every business that relies on equipment (often in remote or inhospitable environments) knows the challenge: sensors produce oceans of numbers, devices spit out cryptic logs, and your teams are left piecing it together under pressure. Machine learning (ML) has been great at crunching the numbers. Now, tiny language models (TLMs) like Gemma 3 270M provide an opportunity to take this one step further, reading the logs, interpreting anomalies, and explain issues in plain language. There appears to be real potential in combing these approaches. (For those more technically inclined I have included a conceptual design and explanation at the end of the article.)

We're all familiar with the massive, powerful language models that run on vast server farms. What if the next big breakthrough in AI isn't about being bigger, but smaller? Over the weekend I fine-tuned Gemma 3 (270M) end-to-end—LoRA → merge → GGUF → Ollama and ran it locally. It wasn’t perfect (tbh, it was more of a learning exerciser to understand the process), but it was fast, inexpensive, and genuinely useful for narrow, domain-specific tasks. Here’s what tiny models are, why they matter to business, and how to get started without boiling the ocean.

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.

As educational establishments seemingly wrestle with how, or if, Generative AI (GenAI) should be formally integrated into their curricula, the conversation seems to circle around a familiar tension: education versus training (I'd love to hear from people embedded in the education space for their opinion). Should STEM degrees remain focused on deep technical foundations, or adapt to include the practical AI skills employers well expect? One promising middle ground is adding humanities courses that sharpen critical thinking, ethics, and communication - capabilities essential for using AI responsibly. The challenge is finding the right balance so educational establishments can preserve their mission to educate while preparing graduates for the realities of an AI-enabled workplace.

Generative AI is becoming a staple in the modern workplace - but something’s not clicking. Despite the rollout of training programs and hands-on tools, it seems that some organisations still struggle to see meaningful impact. Why? Because knowing how to use GenAI isn’t the same as knowing how to work with it. I have been delivering training on GenAI for over a year now and the feature that stands out in the true adopters has been the Exploratory Mindset - it's the mindset that really makes the difference.

You didn’t sign up for an AI platform but suddenly, your HR tool summarises resumes. Your file-sharing service suggests email replies and your CRM is auto-generating forecasts. Welcome to the new world of silent AI rollouts, where vendors quietly add GenAI features to your software stack, often without clear notice, control, or consent. It's not just a tech issue it's a business, legal, and risk management issue.

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.

It happened to me a quite a few years ago when I resurrected some of my C code from the 90's and brought it up to date and if you're a Business Analyst or Developer, you've been there as well: trying to decipher a legacy system with outdated documentation and only a handful of power users to guide you. Traditionally, we've relied on user interviews and painstaking manual testing to map out functionality. Using LLM's, combined with the more traditional methods can give us extra insight.

When it comes to Generative AI, many organisations feel overwhelmed, that they need a massive, enterprise-wide initiative to get started, but you don’t. Whether you're a small-to-medium enterprise (SME) or a single department within a larger organisation, you can begin your GenAI journey with a few focused steps. No massive enterprise-wide rollout project is required - just smart, strategic, thoughtful action in a quick-start approach.

Being immersed in the world of AI can feel like being caught in a whirlwind, every week brings a new model, a fresh feature, or a must-try tool and the pace is not slowing down - it’s easy to get swept up in it all. That’s why I always value conversations with businesses that bring things back to what really matters, solving real problems. GenAI isn’t just the latest shiny object, it’s a powerful tool to unlock capacity and drive real value, when focused on a business problem.

As artificial intelligence continues to reshape industries globally, Canada finds itself at a critical point in AI regulation. With the recent appointment of Evan Solomon as Canada's first-ever Minister of Artificial Intelligence and Digital Innovation, we are seem to be charting a new course that prioritises economic growth while maintaining responsible oversight. Where does this leave organisations that are eager for clarity and direction in an increasingly complex global and asymmetrical regulatory environment?

I've seen it in so many organisations, that growing pile of Should-Do projects gathering dust while teams scramble through endless To-Do lists under the pressure of everyday work. The story is always the same: "We'd love to explore that new product", "We should really improve our staff on-boarding", "We could really improve our end-user experience if only we .....". The constraint? Never enough time, budget, or human intellectual capital. Generative AI is about to hand you back some of that capacity (and capability) - what you do with it will determine whether you're leading your market in the future or wondering what happened to your competitive edge.

AI assistants are now integrated into the data tools you use daily, helping to reveal the stories and patterns in your data with natural language commands. Two of the tools I have been working with lately are Google's 'Analyse with Gemini' in their Colab Notebooks and Microsoft's Copilot in Excel with it's 'Advanced Analysis' prompt. These tools, while on the surface appear similar, do have differences that make them suitable for different types of users.

I wrote about Deep research models back in February and since then I have been actively integrating them into my business processes (both ChatGPT and Gemini version) and I thought the subject was worth revisiting in more depth. For me they have made a huge difference in everything from developing strategic plans to pre-meeting research.

The headlines are buzzing with Machine Learning (ML) and Generative AI (GenAI), and it's easy to think you need a whole new toolkit to stay relevant. While these technologies are transformative, the good news is that many of your existing skills in technology operations are not just relevant, they're crucial. The game is changing, but the foundations you've built are more important than ever. Let's explore how.

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.

Generative AI (GenAI) is promising to reshape how we work and innovate. Yet, many organisations hesitate, wondering, "Is our data perfect enough?" However, perfection often isn't the prerequisite you think it is.

We've all been there, staring at a blank PowerPoint, knowing the information is in a document somewhere, and dreading the hours it'll take to transform it into a compelling presentation. This is where Microsoft Copilot in PowerPoint can help save you hours of time. (At the moment it's my most heavily used feature in the M365 Copilot suite.)

Buying a Generative AI solution, whether it is a discrete or embedded GenAI solution, isn’t like buying a CRM or ERP system. It’s a whole new ballgame, one where you can’t always see the rules, and the players (the models) can sometimes make up their own. GenAI procurement requires a fresh playbook. Let’s break down what’s changing, why it matters, and how you can stay ahead.

As the adoption of Generative AI grows inside the enterprise, automation is entering a new phase, one that blends decision-making, integration, and user interaction in real time. Enter Agent Flows from Microsoft Copilot Studio.

A number of threads are converging: Coding assistants like GitHub Copilot, Codeium, Cursor, and Replit Ghostwriter are integrated directly into IDEs - or in some cases, are the IDEs. Autonomous agents or agent environments that can generate and execute code, like Microsoft Magentic-One or Langflow . Application generation environments like Replit takes this even further, generating full applications from natural language prompts and deploying them with just a few clicks.

Following on from last weeks article on GenAI readiness assessments, to harness its full potential, organisations must adopt a strategic approach. Developing a Generative AI Roadmap is crucial to guide this journey effectively, no only for the GenAI solutions but the development of the systems and processes around GenAI to ensure it's long term support.

Generative AI (GenAI) is revolutionising industries by automating content creation, enhancing decision-making, and sparking innovation. But before jumping in, it’s essential to pause and do the groundwork. In today’s fast-moving, hype-driven tech landscape, that can feel a bit old-school, but understanding your organisation’s readiness isn’t just prudent, it’s foundational. A GenAI Maturity and Readiness Assessment is a crucial step to ensure that the excitement of innovation is backed by the stability of preparation.

Discover how Generative AI can enhance your Business Continuity Planning (BCP). Learn how AI can identify risks, streamline plan reviews, and improve resilience. Embrace AI for smarter, faster decision-making in times of crisis.

Learn how to build a Generative AI portfolio and create strong business cases for AI investments. Discover strategies to maximize ROI, manage risks, and align AI projects with business goals.

Organisations will turn to external partners for GenAI implementations and the success of your GenAI project depends heavily on the clarity of your requirements. A well-prepared RFP with comprehensive requirements isn't just bureaucratic paperwork - it's your blueprint for AI implementation success. Let’s dive into some insights on creating requirements that will attract the right vendors and set your GenAI project up for success.

When discussing GenAI one concern that consistently comes up is a worry around using public models and what happens to my data - there are genuine concerns about data privacy and security when using public AI models.  Luckily there are solutions that address these concerns that have been around for quite a while – Ollama , and Open WebUI - tools that empower organisations to run AI models on their own infrastructure.

Remember when we used to install software from CDs?  Then SaaS arrived and changed everything.  Next, we'll witness an even bigger revolution: AI agents transforming how business SaaS applications work. While Salesforce, Workday, and countless other platforms transformed how we work, their application layer could be replaced by intelligent agents that bypass interfaces altogether to work directly with you and your data.  The applications will change dramatically, but your business data remains the constant foundation of value.  Here's why this seismic shift will reshape every industry and why you need to prepare.

In today's business environment, staying ahead requires not only quick decision-making but also access to comprehensive, reliable information. GenAI Deep Research Models can help - they are AI driven tools designed to transform how businesses conduct in-depth research and analysis.

Before I get started on this one, it's largely a thought exercise on where we could be taking Agents and the impact on the traditional API integration model. The API landscape will shift. Just as APIs revolutionised how applications communicate, AI Agents are likely to emerge as the next solution for system integration. This transformation isn't just about new technology – it's about fundamentally changing how our software systems interact, learn, and evolve together.

In today's fast-paced business environment, Project Managers are continually seeking innovative tools to enhance efficiency and deliver successful outcomes and the Project Management Institute (PMI) seems to be stepping up.

Business Analysts (BA's) are in an amazing spot right now (and in the foreseeable future). GenAI offers unprecedented opportunities to enhance efficiency, drive process improvements, and deliver greater value both within the BA's own processes but also within the processes of the organisation.

I have written previously about technical debt and GenAI choices and when you add on Shadow GenAI the situation becomes even more complex. The accessibility of Generative AI (GenAI) tools means that everyone can innovate like never before. However, this surge in un-monitored GenAI usage, often referred to as "Shadow AI", combined with GenAI only now starting to come down from the 'Peak of Inflated Expectations' can lead to significant challenges, including the accumulation of technical debt within organisations.

Agentic AI seems to be talked about constantly and as I've talked about in previous articles design patterns matter - same thing applies to AI Agents, design matters. What design patterns, or a least design considerations are there for Agent AI design?

In preparing last weeks article I came across a series of blog posts from Ardoq (thanks to Ed Granger) that deserves more attention (catch it here). There are quite a few posts (maybe too many) on how GenAI and Agentic AI can be used in organisations but not much on how we conceptualise and plan for this from a Enterprise Architecture (EA) perspective. Ardoq's insightful blog series delves into GenAI's impact on EA and more - it's really worth a read.

While mulling over what seemed to be a lack of GenAI frameworks and design patterns last week, this post from Debmalya Biswas dropped into my feed on A Comprehensive Guide to Agentic AI. This post and the accompanying slides are more than the title suggests - they are a great start on a set design patterns and reference architectures - something that we seem to be sorely missing.

We've been seeing Agentic AI hit the press a lot over the last while, especially as we head into 2025. I recently had the chance to build a PoC and find out what Agentic AI was all about and I think I get it now...

A little while ago I wrote an article that talked about the urgency of adopting Generative AI (GenAI). The more organisations I speak to, who then see the potential of GenAI, the more I realise that it's an imperative - that the world is changing around organisations and those that are not planning to adopt are leaving both time and cost savings on the table (the ROI is tangible). This hesitation could be costing businesses more than they realise - the loss of a competitive edge.

Agentic AI and LLM tools enable remarkable capabilities, from automating workflows to generating content and insights. As I spend more time in Langflow and really start to appreciate the power of the systems that can be developed I started wondering about how they could be monitored, how do we implement Observability - then a note about Datadog LLM Observability came across my feed and got me thinking that this is worth looking at more deeply.

Mulling over why systems like ChatGPT, Perplexity, Gamma, Gemini, Claude etc. are so successful - it's all about the interface. It isn’t just about their perceived intelligence - it's about how they interact with us - it's about the interface; the natural language interface . It’s a shift in how humans and machines collaborate. What does this mean as an approach to business.

A week or two ago I came across an article from Business Standard that summarised a Gartner report suggesting that over 30% of GenAI projects won't survive beyond proof of concept (PoC) and will be dropped by the end of 2025. Having run a large project portfolio I'm always interested in stats like this so I decided to pick at this a little and see whether this is indeed an issue, or just a headline.

The upcoming rise of Agentic AI - AI that operates more like an independent worker than a traditional tool will fundamentally transform the landscape of work. This evolution prompts an important question: what role do unions play in a world where AI acts as a virtual employee?

The world of AI is evolving rapidly, moving from passive tools to dynamic, "agentic" AI - technology that can operate autonomously, making decisions, interacting with employees, and handling tasks like a true virtual team member. While this shift brings exciting opportunities for efficiency it also brings new challenges for oversight, ethics, and integration into workplace culture. HR stands at the heart of this, ensuring that these “virtual employees” align with company values, policies, and workforce goals.

Am I living in a GenAI echo chamber? While my LinkedIn feed overflows with the latest AI breakthroughs and 'must-try' features, my experience in the trenches tells a different story. As a volunteer leading GenAI projects, delivering prompt engineering training and talking about GenAI in the non-profit sector, I've witnessed a gulf between the breathless pace of AI innovation and how most people actually use these tools day-to-day. (I have to say that I have not noticed resistance, concern - yes, but not resistance and in all cases I see the 'wow' moment happen when people realise the possibilities and practical applications.)

I couldn't believe the timing this week. As I'm working through the course material for the Microsoft AI Engineer exam a post dropped from Rajani Janaki Ram with a Framework on Adopting AI on Azure - brilliant! I thought I'd review the framework this week and what it means.

Generative AI (GenAI) is full of potential, but like any rapidly evolving technology, it carries the risk of technical debt. Making the right decisions now can prevent future headaches. Let's explore how you can be shrewd about your GenAI investments.

After completing certifications in the various platforms from IBM, Amazon Web Services (AWS), Microsoft and Google I wanted to take a high level look at what these platforms really mean for organisations. These organisations are offering powerful AI platforms that provide a way to access AI functionality without having to develop and run your own infrastructure and manage your own AI deployments.

As AI continues to revolutionise industries, understanding and mitigating the security challenges around large language models (LLM's) is critical. The OWASP Top 10 for LLM's is a comprehensive guide to the most pressing risks faced by these models.

For quite some time I have been concerned about the unrealised value in unstructured data - the myriad of Word documents and PDF's that contain everything from organisational policies to processes and reports (we aren't talking about video, images and audio in this article). This increasing amount of unstructured data and the ability to absorb it is one of the things that increases the time that new hires take to become effective or means that a policy (if not encapsulated within a system) does not get adhered to.

(This week is an experiment to see what you think about the new ChatGPT o1-preview model output. I iterated through prompts to ask o1-preview to introduce itself and explain its abilities using my usual article format. The title of this article and all of the main article text below are 100% generated by the o1-preview model. What do you think? My opinion is at the end. For the record it took 29 seconds to generate the content.)

As I explore the AI development platforms from major providers like IBM , Google, Amazon Web Services (AWS) and Microsoft it's clear that they are increasingly offering tools that provide easy access to both pre-trained models and custom model training (we also have easy API access to models like ChatGPT and Gemini). A recent report from the IBM Institute for Business Value included a sentence that resonated with me: "The competitive edge that generative AI delivers today will be table stakes tomorrow." This insight feels particularly relevant when considering the services these major platforms offer - its power to differentiate will diminish.

Understanding the risks in any organisation or project takes time and usually involves one or more risk workshops, more often than not starting with a blank sheet of paper. Massachusetts Institute of Technology (MIT) have provided us with a short cut to identify risks associated with artificial intelligence using a new resource, the AI Risk Repository - save time and improve the breadth and depth of risks.

In niche non-profit grant-making, informed decisions are essential. Relying on accurate data helps align granting strategies with the organisation mission, vision, and goals, as well as addressing community funding gaps.

As organisations strive to scale AI across their operations, a new playbook from Massachusetts Institute of Technology emerges for those ready to bridge the gap between ambition and execution.

I needed a small utility to search using the Google custom search API and then submit the results to ChatGPT's API for summarising; I leveraged the power of the AI coding companion, Codeium for the first time (in VSCode and Python). This combination truly improved my development experience, making it a lot more efficient and enjoyable, auto-completing code, generating explanations and providing in code documentation. Digging a little deeper ...

In a Gartner presentation by Max Goss this week the subject of Organisational Change Management (OCM) in AI projects was brought up. OCM has always been important in IT projects but what makes it different in AI projects? Well, no surprise, it's not just about the technology, it's about the usual suspects such as communications, involving the right stakeholders, leveraging change networks and delivering training but there are more things to consider in the rapidly evolving landscape of AI.

In a significant move to bolster AI governance and compliance, ISACA has unveiled its AI Audit Toolkit. This toolkit is designed to help organisations navigate the complexities of auditing AI systems, providing a structured approach to assess and ensure their AI technologies are both effective and ethically sound. Although focused on Audit, why not use it to help drive design work?

In the wake of the recent CrowdStrike incident it's easy to become an armchair critic. For those with experience in IT, isn't it likely that such issues are multi-dimensional, spanning technical, managerial, cultural, and even simple human errors?

After posting last weeks article (Deploying AI: What nobody seems to be putting together) where I suggested that we have fragmentation in approaches to AI deployment I saw a number of things come together. EY launched EY.ai (actually, back in Sept 2023), then a great post from Chris Howard at Gartner (Top of Mind - "Are 100% Accurate AI Language Models Even Useful?") where Chris mentions '...as we hit the trough of disillusionment with generative AI...".

No surprise – it's a ('it' in this case being the deployment of AI from a corporate perspective) a challenge and an opportunity.  From the Artificial Intelligence perspective, businesses have access to a variety of tools and frameworks designed to facilitate AI adoption....that often operate in silos.

As AI continues to reshape industries and societies, governments worldwide are grappling with how to regulate this transformative technology. Canada has taken a significant step forward with the proposed Artificial Intelligence and Data Act (AIDA). This article investigates what AIDA could mean for businesses and how to prepare for its implementation.

These two subjects, with separate sections below are intertwined - understanding AI risk and the management of AI systems go hand in hand. Luckily NIST and ISO provide us with some tools to help with both.