Getting Real with AI

It has been quite a February for anyone watching the stock market. On the 22nd, James Van Geelen and Alap Shah of Citrini Research published a thought experiment on Substack - a fictional scenario about what might happen if AI displaced enough white-collar workers to hollow out the consumer economy. They explicitly labelled it a scenario, not a prediction. By Monday morning, the Dow had fallen 800 points. [1]

A few days earlier, a product announcement from Anthropic, the company behind Claude, had triggered IBM's worst single-day drop since 2000, wiping $31 billion off its market value in one session. [2] CrowdStrike fell nearly 10%, Cloudflare 8%. Across the software sector as a whole, roughly $2 trillion in market capitalisation has been wiped out since the sector peaked in September 2025, and hedge funds have made $24 billion shorting the sector in 2026 alone. [3]

Wall Street, it is fair to say, does not yet know how to price AI.

And yet, whilst the markets panic, the practitioners are getting on with it. According to the Clio 2025 Legal Trends Report, 87% of professionals in large law firms already use AI in their daily work. [4] Not experimenting with it. Using it. The gap between the headlines and the reality has probably never been wider.

That gap is what this series is about.

Most people's experience of AI so far is a chatbot. You type a question, you get an answer, and then you carry on with your work. The AI sits beside you as a useful assistant, certainly, but you are still doing the work. Your processes have not changed. Your workflow is the same as it was. You are just a bit faster at some parts of it, which is welcome, but it is not really what the fuss is about.

The shift that matters is when AI does a task, not merely assists with one.

• When you automate a workflow, rather than speed up a step within it.

• When an analysis that took your team a day happens overnight, every night, without being asked.

• When your systems learn from what you teach them and quietly apply that learning going forward.

That is the difference between a slightly better search engine and a genuine change in what your team is capable of.

So how do you get from chatbot to capability?

In our experience, three things make the difference - and none of them starts with technology.

First, understand your business problem clearly before you go anywhere near an AI tool. Not 'we should use AI', but 'what specifically is costing us time, or money, or accuracy?' Where are people doing repetitive work that a machine could handle? Where is data sitting in one system that should be informing decisions in another? The problem comes first. The technology comes second.

Second, be realistic about what AI can actually do right now. Not the hype, not the doomsday scenarios, but the practical capabilities available today on the hardware you can access — whether corporate or consumer - for a modest investment. AI is very good at pattern recognition, classification, summarisation, and generating structured output from unstructured input. It is not good at judgment, strategy, or knowing your business context. That part is still yours.

Third, apply a simple discipline. Every use case we have built follows the same five steps: define the business need, define the outcome, review the plan, delegate the tasks, and review the outcome. That is not a technology framework. It is a management framework - the same process you would use if you were delegating to a capable but new member of staff. Which, in a practical sense, is exactly what you are doing.

I think of AI as a highly intelligent person who has spent forty years studying in a Further Education college - TVET, for those of you in most of the world - and knows how to do almost everything remarkably well. The problem is that they have never actually worked. They have no experience of consequences, no context for your business, and no real sense of what they do not know. The world's best and worst intern.

One thing worth saying clearly: this is not about which model you use. In 2026, the leading AI models are all very capable, and the differences between them are more about working style than raw ability. Some tools suit people who think in conversation; some suit developers who want to define a task and walk away. Some jobs are best served by small, local models running on your own hardware. We have processed 50,000 photographs on a standard office machine using local models for image indexing and website accessibility text, with no cloud dependency, no subscription, and no specialist equipment. The point is not the tool. It is what you do with it.

My colleague Viren Lall recently published a six-part series on leading AI in organisations, covering the strategic questions: should we adopt AI, how do we lead the change, and what does it mean for our people? This series is the practical companion. Over the next four articles, I am going to walk through four real use cases that we have built and are using at ChangeSchool every day.

• Using your CRM data as a back-end and building your own interface on top of it — one that actually maps to your sales process and your people.

• An email triage system that learns how you work and quietly sorts your inbox so you only see what matters.

• A tender and proposal writing system powered by hundreds of your own documents, searchable by AI on your own hardware, producing grounded and evidence-based responses.

• Personal software: command-line tools built in an hour using publicly available APIs, that do exactly what you need and nothing more.

These are real projects with real results, and I will be honest about what worked and what did not. If you are a business leader wondering how to move beyond chatbots and make AI work in your organisation, this is where to start.

Sources

1. Van Geelen, J. and Shah, A. (Citrini Research), 'The 2028 Global Intelligence Crisis', Substack, 22 February 2026. Market reaction reported by Motley Fool and Bloomberg.

2. IBM stock drop: Bloomberg and CNBC, 23 February 2026. Anthropic Claude Code COBOL modernisation announcement, 23 February 2026.

3. $24 billion short-seller profits: CNBC, 4 February 2026, citing S3 Partners data. $2 trillion market capitalisation loss: reflects the drawdown from the sector's September 2025 peak across multiple sources. IGV ETF down approximately 23% year-to-date as of late February 2026.

4. Clio Legal Trends Report 2025. The 87% figure refers to professionals in large law firms.

This article is part of The Practical AI Series on practical AI applications at ChangeSchool. For the strategic view on leading AI in organisations, see Viren Lall's companion series. ChangeSchool is an SME without a dedicated coding resource. Every application described in this series was built using AI coding tools — all within the reach of a start-up or micro business willing to invest some hours and have a go.