The agents
Systems, not prompts.
Three AI agents I built and run — first for HWM Global, the water-technology brand I market to utilities worldwide through a network of regional distributors. Each takes a goal, breaks it into steps, and uses tools to do the work — agentic, not a single prompt for a single post. Each gives one person a running start on work a whole team would normally carry.
The bar has moved from "I use AI" to "look at the judgment behind how I use it." So these aren't demos of content generation — they're working tools, with the decisions I made (and the things I deliberately left for a human) on show. They're gated by a token because every run is real API spend; email me for one.
Agent 01 — Brand voice
Brand Voice.
On-brand by construction — across every platform, in every market a distributor sells in.
It reads HWM Global's brand guidelines and encodes them into a reusable Voice Profile — voice, tone, lexicon, prohibited claims, visual rules. Then anyone in the company, or any distributor anywhere in the world, can hand it an asset and a market and get on-brand copy for the platform they need — social, web, email, a case study, a video script, an event — tailored to how that region actually buys. Not translation: cultural and tonal re-tailoring, with the reasoning shown. Every version then passes a governance check — brand-voice flags, prohibited-claim hits, and a queue of what a human must approve before it ships.
The judgment: the prohibited-claims gate and the human-approval queue are the point — they turn AI speed into consistency at scale rather than inconsistency at scale, and keep a person on the high-resonance, high-risk calls.
How it works
Agent 02 — Product launch
Campaign Orchestrator.
One product-launch goal in; a 360° campaign — researched, positioned, drafted, QA'd — out.
Give it a new product and a market and it runs the whole launch: market and competitor research, positioning and naming, the messaging hierarchy and channel mix, then the actual assets — social, email, landing page, trade press, event — each drafted and run through an automated brand-voice check. The clearest example of agentic-not-generative: it plans the launch and executes it end to end, not one post at a time.
The judgment: it's a running start, not a rubber stamp — the price and accuracy claims, the competitor comparisons and the regulatory framing are surfaced for a human to sign off before anything ships.
How it works
Agent 03 — Search & AI visibility
SEO + GEO.
Show up when buyers search — and when they ask an AI.
Search is splitting in two: the ten blue links, and the answer an AI gives. This agent takes the brand and its industry, generates the questions a real buyer would ask, then puts each one to ChatGPT, Gemini and Claude on the live web — capturing what each engine says, the sentiment toward the brand, the competitors it names instead, and the sources it cites. It ends with a cross-engine read plus the SEO keywords to own and the citations worth earning to move both the search result and the AI answer.
The judgment: the work isn't the scrape — it's knowing which questions matter in your industry, reading where the brand is invisible or losing, and deciding which keywords and citations are worth going to earn. It queries all three engines — ChatGPT, Gemini and Claude — side by side and analyses each.
How it works
All three were built with Claude Code — agentic developer tooling, used to ship working software, not just content.