This morning, I sat down with Doyl Burkett, Managing Partner of Integrity Growth Partners, a Los Angeles-based growth private equity firm with over $800M in assets under management. What makes Doyl unusual is that most GPs are focused on what AI means for the companies they invest in. Doyl is equally focused on what it means for the firm itself. The lazy version of where we are is that AI breaks PE. Doyl's view is more nuanced. He thinks AI isn't going to compress the asset class evenly, it's going to sort it. Companies will end up in one of three buckets, and the same sorting is going to happen to GPs themselves. He's running that experiment right now at Integrity. He calls the framework CAPE, and it runs across sourcing, reporting, and operations. Some of the highlights: ↳ AI won't break PE — it'll sort it. The hot take is that AI compresses the whole asset class evenly, a kind of peanut-butter spread of doom. Doyl's view is the opposite. It widens the gap. Companies fall into three buckets — AI-native, defensible-moat, and undifferentiated — and the distance between them only grows from here. ↳ The same sorting is coming for the GPs themselves. The take he said would start an argument at a GP dinner: firms should be using AI as aggressively as the tech businesses they invest in. Most nod, then go back to doing things exactly the same way. He thinks that gap eventually shows up in returns, and in who can still raise a fund. ↳ He ate his own dogfood in the best possible way. A prop-tech portfolio company needed to build AI to cut out the third parties it was splitting revenue with. Mid-build, the CTO said an acquisition could short-circuit the whole thing, so they pointed their own AI sourcing engine at finding the target, bought a tiny AI-native company, and integrated it. That single move is taking ~$11M of revenue toward nearly double. ↳ Be very suspicious of the AI quick fix. His analogy: most "AI for sourcing" or "AI for ops" tools are a Ferrari built for an F1 driver, when what the business actually needs is a Ford Focus. He's spent years and millions building CAPE precisely because there was no shortcut, and he'd warn a friend off believing there is one. Full episode dropping soon.

The AI market in 2026: capability-rich, pattern-poor. I've spent two years asking applied AI experts the same set of questions. There's no shortage of "what's possible" content. There's a real shortage of first-hand accounts: structured, searchable, written by the team that built the thing, not the team that sold it. Today's issue (first of the new format focused exclusively on AI use cases and case studies) features two builds worth pulling up: Case 1: a PE-backed e-commerce co replaced its BPO with a multi-model AI pipeline (OpenAI Gemini Claude on AWS). 84% cost cut. 24h → 30s per batch. They built the eval system first, proved accuracy beat the BPO baseline before go-live. Chris Taylor, @FractionalAI (just acquired by the new @AnthropicAI -backed services venture). Case 2: a mid-market PE firm encoded six senior MDs as queryable AI personas via structured interviews on past deals. 5x more analytical angles per IC. 1,000 deals queryable. The scaling pattern: one MD first, nail it, then expand to five more for sector coverage. Osman Ghandour, Soal Labs. Plus my podcast with Tom Scott (CEO of @wrike , $250M ARR work management platform) on why most SaaS CEOs are sequencing their AI rollouts wrong. Real builds, attached to real people you can call. Every issue from here on: a handful of fresh case studies from the Pluris AI Use Case Library, with a one-click path to the expert behind each. just-curious-ai.beehiiv.com/…

Fractional AI was acquired last week by the new Anthropic-, Blackstone-, and Hellman & Friedman-backed enterprise services venture. Here's an example of the kind of work that put them in that conversation. A PE-backed e-commerce company was hostage to its BPO vendor. Shopping lists arrived as images, PDFs, spreadsheets. All of it had to be matched to the company's internal product taxonomy. The offshore team held all the institutional knowledge. Turnaround was 24 hours per batch. Fractional replaced the entire line with a multi-model AI pipeline. OpenAI for classification, Gemini for text extraction, Claude for intermediate steps, all on Amazon Web Services (AWS). Result: 84% cost cut. 24 hours → 30 seconds per batch. Accuracy above the human baseline. The discipline most teams skip: Fractional built the evaluation system before the pipeline. That meant they could prove accuracy in blind side-by-side tests against the BPO baseline before go-live. Two things came out of that: 1/ They could decommission the BPO with confidence. 2/ They discovered the BPO had been performing worse than anyone assumed. Invisible until somebody finally measured it. Full case study in the comments. I write about applied AI weekly at Just Curious; case studies from the team that built it, not the team that sold it. Check out the case study and all the others in our AI Use Case Library: checkpluris.com/case-study/a…