AI Manifesto: Stage 1, Clarify
Before You Buy the Tool, Name the Job
The fastest way to waste money on AI is to treat it like a category, not a capability.
Plenty of companies are doing it right now. They announce an “AI programme”, assemble a steering group, select a platform, and begin the internal rollout choreography. It feels decisive, but it also puts the cart before the horse. The real work of adoption starts long before the first tool demonstration.
Stage 1 of the Human-First AI Adoption Framework is blunt about that. It asks a question that has nothing to do with models: “Why would AI actually matter to our business?” The discipline demands a specific problem worth solving, anchored in business need, with an early definition of what “good” looks like so it can be measured later. It also insists on a sentence that should be read aloud in any board discussion: AI is a tool in service of your mission, not the mission itself.
The manifesto makes the same point, with a succinct “Purpose Before Adoption.” Deploy AI only where it creates clear value, supporting the human endeavour which delivers on your goals. It is a rebuttal to the wider mood captured in Transvault’s framing piece about the race to AGI, where capability is accelerating faster than wisdom and control.
Why standards come before automation
For practical comparison that gets straight to the point, look at Toyota. The Toyota Production System (TPS) is routinely described as a management philosophy and a socio-technical system, not a bag of tools. Its reputation was built on a discipline that starts with purpose and process, then earns the right to automate. That is why Taiichi Ohno, one of the architects of TPS, is often paraphrased this way: “Where there is no standard, there can be no improvement”.
That line matters for AI, because Stage 1 establishes the operational standard not just the technical one. What exactly are we trying to improve? By how much? With what safeguards? Who signs it off?
Finding the problem worth solving
A good Stage 1 conversation sounds less like a technology briefing and more like a production engineer walking the floor. Where is the time being lost? Where is the rework coming from? Which parts of the workflow are repetitive and low-stakes, and which parts are judgement-heavy? The framework’s guardrail captures it in plain language: “If you can’t name a real, measurable problem, stop here.”
This is also where you start building credibility with the people who will live with the change. John Kotter, whose work on transformation is still the default reference point in many boardrooms, has long argued that leaders reduce resistance by educating people about the need for change and the logic behind it. This phase is where that logic gets written down in human terms, not vendor terms.
The discipline of being selective
The temptation, especially in knowledge work, is to select “use cases” because they demo well. “Drafting emails faster” is easy to show. “Summarising meetings” gets applause. None of that is wrong, but it can leave you with activity instead of positive impact.
A more Toyota-like approach starts by defining the job to be done. Maybe your procurement team is drowning in contract turnaround time and missing savings opportunities. Maybe customer support is spending too long finding the right policy answer, leading to inconsistent outcomes. Maybe compliance review is labouring under volume that is better handled with triage and retrieval, so experts can focus on the calls that require judgement.
This is where Porter’s thinking becomes useful., Michael Porter’s famous line from What Is Strategy? is that the essence of strategy is choosing what not to do. That is the discipline again. It is permission to be selective. If your AI agenda is a shopping list, you have not clarified your priorities. You have collected ideas.
Request the full AI Manifesto
The measurement question nobody asks early enough
There’s another credibility point hidden in the measurement question. People love to quote “what gets measured gets managed”, but it is widely disputed as a Drucker quote and often used as a blunt instrument. Stage 1 takes a better stance: establish what “good” looks like so you can measure it later, while remembering that numbers never remove accountability.
That accountability thread is the foundation for every later stage. The AI Adoption Manifesto says “Humans Decide. AI Assists.” This process is where you define which decisions remain human, and which tasks are safe to assist with. It is also where you set the conditions for “Responsibility Has A Name”, because you cannot assign a true owner to a system you cannot describe in outcome terms.
How Stage 1 earns the rest of the journey
When done well, Stage 1 becomes the organising principle for the rest of the readiness journey. Map, the following stage, becomes easier because you know the destination. Commit becomes grounded because there is a business case, not a vibe. Pilot becomes safer because success criteria exist before anyone builds. Measure becomes honest because it compares reality to a definition agreed earlier.
Toyota did not win by buying the best machines first. It won by relentlessly defining the work, standardising it, and improving it, with humans responsible for the system. Effective AI adoption starts the same way: define work, standardise it, then improve it with humans accountable for the outcome.
Author: Darwin Lee
Request the full AI Manifesto