Action
The craft loop
A six-phase workflow for producing complex, high-quality artefacts in a single working session — when domain expertise and AI capability operate in genuine tandem.
There is a version of AI-assisted work that is fast. You prompt, you accept, you publish. The speed is real. The depth is not.
There is another version. Not faster in name only, but genuinely productive because the human and the AI are working together in a loop that neither could sustain alone. Today I produced a fully branded, interactive digital presentation, complete with keyboard and mouse navigation, responsive mobile design, and AI-generated voiceover narration on every slide. Design agency territory. The kind of brief that gets quoted in weeks. We completed it in a day.
Not because the tools are magical. Because the process was right.
The shape of the workflow
The process follows a clear sequence: brief, build, challenge, enrich, adversarial review, publish.
Brief is where the SME sets intent. Not just "make me a presentation." Intent means: audience, emotional register, what the artefact must do, and what it absolutely must not. The brief going in shapes everything coming out.
Build is where the AI generates the base artefact. Quickly. Broadly. Not finished. A high-quality draft that gives the SME something real to react against. The AI is not trying to be right here. It is trying to be useful. A good initial draft is provocative, not polished.
Challenge is where the quality lives. The SME brings domain knowledge to bear against the output. Every framing gets tested. Every sequence questioned. Narration scripts get rewritten until they sound like a person thinking, not a model performing. This is not editing. It is intellectual tension. That tension is what separates generated from good.
Enrich is where the artefact gains context and layers. Brand integration, visual identity, interactivity, audio. Functional value dressed for the room it will walk into.
Adversarial review completes the loop. Using a competing AI system surfaces gaps that a single model family would miss. The two systems have different training distributions and therefore different blind spots. Running both in sequence means those differences work in your favour, not against you.
Publish is a commitment, not a step. The work carries the SME's name and reputation. That accountability disciplines every phase before it.
What makes it work
Three things, and only three.
The SME must be genuinely present at every stage. Not delegating quality. Participating in it. The AI brings velocity, structure, and production reach. The human brings direction, judgement, and accountability. Neither can hold those roles for the other.
The adversarial quality loop cannot be skipped. Two AI systems applying pressure from different training perspectives catches more than any single pass.
The brief cannot be abbreviated. Vague intent produces coherent output that misses the point. Every minute invested in a precise brief comes back as reduced rework.
The foundation beneath the loop
There is a precondition for this workflow that the diagram does not show: the AI must already know the person using it.
The quality of the brief, the precision of the challenge, the accuracy of the enrichment: all of these depend on accumulated context. Writing patterns, domain constraints, communication preferences, project history. The kind of knowledge that usually lives only in someone's head, or in years of working relationship.
When that context is invested deliberately, built and maintained as a foundation rather than accumulated by accident, the craft loop runs faster and closer to the mark from the first iteration. The AI already knows what good looks like for this specific person. That changes what the challenge loop costs, and what it catches.
This is not a feature of any particular AI tool. It is a function of how seriously the practitioner has taken the relationship. The loop is the visible part of the work. The foundation is the invisible multiplier.
What this is not
This workflow earns its cost when the artefact is complex, multi-layered, time-pressured, and demands both domain depth and production quality simultaneously. A quick LinkedIn post does not need this process. A conference presentation for a specific audience, at a specific moment, making a specific argument: that does.
The thing worth saying
We are still learning what genuine co-creation looks like in practice. Most conversations about AI productivity focus on the tool. The actual constraint is the human: how precisely they can articulate intent, how rigorously they challenge output, how honestly they hold the adversarial loop.
The ceiling is not where most people think it is. Getting there requires showing up as a craft partner, not a prompter.