I started building AI into Metamorph last year. Plugged Claude Code into a YouTube tutorial, wired a few skills together, watched the first workflow run on a live search. I genuinely thought, "this is going to be easy."
I was wrong about everything that came after.
Not wrong about the destination — I knew where I wanted the firm to land. Wrong about the route. The iteration depth. The assumptions I'd baked in without realising. The ways a workflow that worked for one search type fell over on the next.
What I planned vs. what I shipped.
Take one example: BD intelligence.
The plan was simple. Scan LinkedIn for newly-appointed CEOs, backfill their previous companies, reach out fast — the kind of work that pays off when you catch someone in the first 90 days of a new role. Work I was already doing manually, badly, when I had the time.
What I shipped was six specialised skills and several sub-agents working in concert.
Why? Because once I started building, I noticed the gap I'd been ignoring: LinkedIn data is too late. Most people don't update their profile until they've been in the role for weeks. So the build pivoted to predictive signals — leading indicators that someone was about to move. That's not one workflow. That's a stack of small, specialised agents, each watching one signal, feeding the system.
The part you don't read in the deck: building reveals what using-the-tool hides.
Sit with a workflow long enough and you find the assumption you didn't know you were making. Then you fix it. Then you find the next one.
I love this part. I also underestimated it by an order of magnitude.
The moment it stopped being about my firm.
Six months in, I plugged everything into Obsidian as a second brain — CRM data, brand voice, ICP definitions, every previous brief. Claude could now pull from a structured memory of my business, not the open web.
Then, one weekend, I downloaded a website-build skill from a YouTube video, plugged it into Claude with my Obsidian context, and built the site you're reading. In about two hours.
Around the same time, I was shipping new Metamorph skills every week. I started talking to other recruitment founders about AI, and I noticed something. Most of the conversation was about existing tools that claim to have AI. ATS vendors adding a chatbot. Sourcing platforms with a smart filter.
What I was doing wasn't bolting AI onto a tool. It was building tools around my philosophy.
That's when it stopped being about my firm. If I could do this in six months on my own, every small recruitment firm in the world could do it.
Build vs. buy: it's not what you think.
The argument isn't "build is better than buy." Tools earn their place. The argument is more uncomfortable:
When you buy a tool, you conform to its view of recruitment. Its workflow. Its data model. Its assumptions about how a search should run.
When you build, the system conforms to your philosophy. How you screen. How you brief. How you carry context from one search to the next.
In a market where every recruiter has the same job boards and the same LinkedIn licence, your workflows thinking the way you think is the differentiation.
Not faster. Not better tech. Yours.
What it actually costs.
The full Metamorph stack — Attio, Obsidian, n8n, Clay, Claude Max, Airtable, Unipile, Instantly — runs me under $500/month and does the work that used to take an entire research function. Most tools are $10–$30/mo; Claude Max is the biggest line at about $100. Legacy tools I replaced (GoHighLevel, Otter, the old host) covered half the new spend.
One lesson the hard way: when testing a new tool, stay on the free plan and rate-limit yourself. I burned a month of Firecrawl credits in five minutes by forgetting to set scrape limits. That kind of mistake compounds across a stack.
If you're considering AI help, ask two questions.
First: "What's your actual experience running searches?" Most AI consultants are technologists, not recruiters. Smart people, capable people — but they haven't lived the administrative hell of a live retained search. Without that context, the workflows they design solve the wrong problems.
Second: "What's the single biggest impact area for my firm right now?" A good answer starts with an audit, not a feature list. Then a low-effort/high-impact pilot. Then build out from there.
If they answer with a tool, they're selling. If they answer with a process, they've thought about your business.
What to do this week.
Sit down with a big cup of coffee. Forget the AI question for a second.
Where do you want your firm to be in five years? Not the comfortable answer — the honest one. Then work backwards. What processes does that firm need? What's the gap between those processes and what you're running today? Where does technology fit into closing the gap?
Pick ONE thing. The single most impactful workflow. Treat it like the first build, not the last.
Or just book a Strategy Call. We can do the coffee bit together.
The Customer Zero case study
The stack, the modules, the metrics, the painful bits.