Case study
AI workflows - useful automation, not magic
I got bored of doing the same content tasks manually, so I started turning them into pipelines. The tools I've built at Beaphar and the content systems at Releaf came out of this approach: find the repeatable bit, automate it carefully, keep a human on the decisions.
The brief
- Goal: Stop doing the same thing manually every time - content briefs, social copy, reporting pulls, competitor checks. If the steps are the same, it should be a pipeline.
- Constraint: No "magic box" outputs. Everything needs guardrails, compliance checks, and a human making the final call.
- Principle: AI works best when it’s part of a system - templates, checks, versioning and feedback loops. Not when you paste into ChatGPT and hope for the best.
How I built it
- Prompt to pipeline: A prompt that works once is nice. A prompt wrapped in templates, validation and output processing is a tool. That's the difference between “using AI” and building with it.
- Quality gates: Every pipeline has checkpoints - compliance checklists, tone rubrics, stop conditions for when outputs aren't reliable enough. The Social Post Workshop bakes 650 words of brand rules directly into the system prompt.
- Human in the loop: AI does the repetitive bits. Humans make the calls that matter - tone, accuracy, whether something should ship. No exceptions.
- Safe experimentation: Test locally, version changes, roll back when things don't hold up. The reporting pipeline at Beaphar started as a rough script and got refined over several iterations.
What shipped
Brand-enforced content tools
Tools like the Social Post Workshop that generate on-brand content through structured workflows - not free-text prompting. Brand voice rules, compliance checks and output sanitisation built in.
Automated reporting
A Claude Code skill that pulls data from multiple sources, structures it, and generates consistent reports. What used to take hours now runs with one command.
AI-assisted content systems
At Releaf, built AI-assisted workflows for keyword research, drafting support and competitor analysis - with compliance QA before anything went live. Helped grow organic traffic 228%.
Monitoring workflows
Automations that watch for changes - stock levels, dead links, data shifts - and alert the team instead of waiting for someone to notice manually.
Results & learning
- Speed with control: Content that used to take a full day now takes an hour - and passes compliance first time because the rules are baked into the tools, not left to memory.
- Consistency: Reusable templates and pipelines mean the output quality doesn't depend on who's using the tool or how tired they are.
- Adoption: Tools that fit existing workflows actually get used. The Social Post Workshop has a three-step flow because that's how the team already thinks about content.
Tools & systems
AI: Claude API, Claude Code skills, Zapier agent automations
Ops: Prompt templates, compliance checklists, QA rubrics, experiment logs
Delivery: Human review on every output, versioned changes, clear rollbacks when experiments don’t hold up
Want to see this in practice?
The Beaphar AI tools case study shows three of these workflows in detail - including a live tool you can try yourself.
