I am no longer an LP.

But like a lot of LPs, when AI got good enough, I had a thought: there must be a better way to do the grunt work I used to do.

Like a lot of LPs, I was mostly disappointed.

I tried ChatGPT on a question I'd asked a hundred times as an LP: who are the PE firms in Denmark? The answer came back confident, fluent, and wrong in a specific way I couldn't immediately name.

The old tools weren't better. PitchBook's filters were a maze - twenty checkboxes deep into the UI before I realised the one I actually wanted didn't exist. Preqin was the same shape of problem, different shade of grey. The hard way - Google searches, spreadsheets, peer calls, hours of cross-checking - built better maps but took weeks. AI was supposed to be the upgrade. It wasn't.

The answer was a list of names - some in Copenhagen, some in Stockholm, some in London with a Danish deal team. Reasonable enough. But useless.

There isn't one question hiding in "who are the PE firms in Denmark." There are at least two.

The first is: do I want Danish exposure in my portfolio? Which firms will deliver that? Under that question, a Stockholm-based firm that does many deals in Danish companies absolutely belongs in the list. Where the firm sits doesn't matter; where the capital lands does.

The second is: I'm planning a trip to Copenhagen. Who do I want to meet? Under that question, the same Stockholm firm doesn't belong at all. I'd have to fly to Stockholm to meet them.

Same words. Two different lists. Almost no overlap.

The same trick runs on any market.

Every question of this shape carries an unstated assumption that quietly determines half the answer before research begins.

The reason ChatGPT couldn't answer the question I asked is that I'd been asking it wrong for ten years too. PitchBook, Preqin, my own spreadsheets - all of them answered the question I typed in, not the question I meant.

I once missed BV Investment Partners for years because my team had labeled them venture in our internal map. A Boston tech-services buyout firm with a $2.4B fund. The kind of mistake that quietly stays in your head until it costs you a meeting.

The methodology was in my head - scoping I did before any market research, things I checked, firms I knew to exclude, sub-strategies I looked for inside large-cap brands. I never wrote any of it down. Nobody does. It's the kind of thing you accumulate one mistake at a time and carry as taste.

So a few weekends ago I wrote it down. All of it. Then I taught Claude to ask the questions before doing any research.

The thing that came out of that is a Claude skill called pe-market-mapping.

A Claude skill is - short version - a recipe Claude loads before doing the work. Not a prompt I remember to paste. Not a clever instruction I hope it follows. A piece of methodology that lives inside the tool and shapes how it behaves from the first message. I ask "map the Danish PE market," and instead of producing a list, it asks: exposure or coverage trip? Which boundary - Copenhagen, Greater Copenhagen, all Denmark, all Nordic? Buyout only or all of PE? What size band? Then it researches against that scope, not against the wrong one.

I ran it on Greater Boston last week. Core mid-market buyout, coverage-trip scope - the map an LP would build before a meeting swing.

Four firms came back in the included list: BV Investment Partners (the one I'd missed for years), ABRY Partners, Charlesbank Capital Partners, Webster Equity Partners. Locked schema on each - HQ, managing partners, latest fund, AUM, strategy, deal size, recent activity, LP base, placement agent, notes, website. Same eleven fields every time. Not "what was easy to find" - what I actually need to know.

Nine large-cap Boston names checked for sub-strategies - Audax, Bain, Berkshire, Charlesbank, Parthenon, Great Hill, TA, THL, Advent. The skill flagged the sub-strategies most "headline name" searches miss: Audax Origins inside Audax Group, BCTO inside Bain, the Automation Fund inside THL. One qualified outright - Charlesbank, via their $1.275B tech-opportunities fund. The others landed in the exclusion file with the specific fund size that disqualified them and what would change it. Origins II closing above $1B would bring Audax into scope.

Thirty-plus named firms in the exclusion file overall. Each categorised - flagship too large, fund too small, wrong strategy, wrong firm type, just outside the geographic boundary, insufficient deal-team presence. Each with the reason and what would change it. And many of these are names I'd want to meet on the trip anyway - strong firms that just don't fit this specific scope. The exclusion file is also an outreach long-list, and because the reasoning sits on the page, I can validate every call myself.

AI organises; I judge.

A sortable spreadsheet sits alongside the markdown for when I want to filter rather than read.

Better than any map I built in ten years with PitchBook and Preqin. Not because the firms are different - because the scope was right from the first question.

Ask anyone for the headline mid-market firms in Boston and you'll get Berkshire, Audax, Charlesbank, Great Hill. Every time. Three of those four land in the excluded file. Charlesbank makes it in - but only via their tech-opportunities sub-strategy. Their flagship is excluded too. Same firm, two faces. Same market, two different maps. The default answer is wrong because the default question is.

The pe-market-mapping skill is free to download at https://fundframe.io/skills. I'll keep updating it - that link will always point to the latest version.

It's the first in a series. More are coming: GP leaver analysis, capital account data extraction, monitoring updates. Boring? Yes. That's where the LP hours go. That's what should be automated first.

FundFrame is what I'm building now. The skill is one piece of how I think the work should be done.

There's something strange about writing methodology for a job I no longer do. I spent ten years doing it without the methodology written down. So it's overdue.

Is it perfect? Most definitely not. The Boston run missed Cove Hill Partners - a Boston long-hold firm with a fund that fits the core mid-market band. I'd catch that on review - the skill should catch it first pass. I'll fix it. Every map sharpens the methodology. Every edge case feeds back in.

But what used to take ten hours of clunky filters and half-right answers is a few questions and a few minutes. It works now. Not in some future version. Not after the next release.

The work was never about getting the answer. It was about asking the question well enough that the answer was useful.

I just stopped keeping that part in my head.

Last Sunday, late evening, wife and toddler sound asleep. I stayed up to watch Barcelona vs. Real Madrid on the iPad beside me.

That also meant I had a few hours and I wanted to use them to upgrade the business.

So I opened the editor and started writing the methodology for a PE market-mapping tool.

This is the constant founder tension. There's the work that feels like progress - the new skill, the cleaner CRM setup, the AI agent that runs while I sleep. And there's the work that actually moves the business - picking up the phone, sending the email, having the conversation I've been avoiding. Both are real work. Only one of them shows up in next month's numbers.

I tell myself optimization is leverage. Sometimes it is. The pe-market-mapping skill is in the main piece above this one, and yes - that's marketing.

But there's a tell. After I finish a piece of optimization, I notice what I do with the saved hour. If I pick up the phone, it was leverage. If I open the editor and start writing another skill, it was avoidance. I'd put myself at 50/50.

The annoying thing is the avoidance feels exactly like the leverage. Same flow state. Same sense of progress. Same satisfaction at the end of the day. The difference shows up two weeks later, in the pipeline.

Making Skills is honest work. So is sales. The trouble is if I only reach for one.