Limble

Limble makes maintenance software. It's the system that tells a technician which machine needs them next and keeps the record once it's fixed. Their CMO reached out to me in his second week on the job with a big idea. He wanted a "marketing brain," one AI system that knew everything his team knew.

I told him yes, eventually. A brain needs something to connect, and most of what his team knew was still living in people's heads. Everyone was already using AI, each in their own way, and nobody was measuring any of it.

On top of that, Limble had just rebranded. The old logo and company description were still sitting on review sites, directories, partner pages, and app listings, which are the same pages Google and the AI answer engines read when a buyer asks what Limble is. And he knew the board would want to see what AI was doing for marketing, in numbers.

Before I built a single thing, I trained the whole marketing team on giving AI the context it needs, so they'd stop starting from zero every time. Two weeks later, their hackathon projects were running on context files they'd written themselves.

Then I wrote the Limble master context file, the single source of truth for the brand. It held who Limble sells to, how it's positioned, what it never claims, and how it sounds, and every agent I built after that read from it.

Two rules held the whole way through. Everything got built inside Limble's own Claude and handed to a named owner on his team, so the work stayed with Limble after I left. And every audit opened with a baseline number and a date, so each rerun gave him a before-and-after he could put in front of his board.

I also held one build back on purpose. The website auditor sat on the shelf until Limble's platform story settled, and once it did, it went from approved scope to scoring pages in nine days.

The first thing I went after was the public record. That's where every buyer starts now, and every AI model too, and Limble's was still covered in the old brand. So I built three agents that hunted down the old logo and the old company description anywhere they still lived on the web, and handed each fix to someone on the team. The next time Limble refreshed the brand, they could just run them again. I also got Limble a proper entry in the knowledge graph the AI models pull from.

Once the brand was straight, I wanted to know whether the story held up. One agent checked every competitor claim against public data each month, so product marketing heard about a shaky claim before a rep got hit with it on a call. Another mined sales calls for how buyers talk about the problem and held that up against the context file.

Then the website, because that's where all of it lands. I built an agent that scored every key page on one question (does this page do its job?) using the claims and competitors the CMO had signed off on, and every page owner got ranked fixes they could paste straight in. The best find was that most of the gaps lived in a few shared templates, so a handful of fixes could move hundreds of pages at once. Two more agents rounded it out. One tracked where Limble showed up in AI answers, and a content librarian indexed every public page, built to get sales from a 30-to-60-minute hunt for the right customer story down to under two.

Video was the fun part. Once the story was settled, I made four 60-second AI films held to Limble's brand bar, including one that turned a /learn article into an explainer the rest of the library could copy. The first one took me a day. Five days later the CMO had sent it to his marketing leadership team.

So by the end, the marketing brain he'd asked me for in his second week was eight agents running off one context file, with someone on his team owning every one of them.