What the Vertical Software Selloff Means for Private Equity AI Playbooks

Almost $1 trillion wiped from software stocks in two weeks. Anthropic released a set of industry-specific plugins for legal, sales, finance and customer support, and the market repriced the entire sector.

I've spent the last couple of years building AI inside PE-backed software companies. Hands-on, inside the teams, shipping products and transforming operations.

Since the selloff, I've been in back-to-back meetings with operating partners and value creation teams. Same question in every room: what should we do differently?

Here's what I see that separates the firms pulling ahead from the ones that aren't.

We'll cover:

• Why the real gap isn't between action and inaction
• The four loops I see driving that gap across portfolios right now
• What it takes to create value in a market where hold periods are extending
• A practical framework for finding what's blocking progress

The Widening Gap

Almost every PE-backed software company I work with has done something on AI by now. Engineers coding with AI. They ran a pilot. Maybe rolled out a chatbot. Or an LLM-workflow feature that they call an agent.

That's not the gap.

The gap shows up in the way these initiatives have been approached:

• They started a pilot without assessing data and legacy architecture readiness, so the pilot never launched.
• They skipped customer validation for their AI features, and after seven-figure investments they see no usage, no results.
• They vibe-coded AI features that worked in demo but are now failing in production and eroding customer trust.

Every one of these was treated as a one-off activity. And one-off activities have a shelf life. The pilot that didn't launch six months ago would need to be rebuilt today because the models have moved on.

The companies pulling ahead are built for iteration and compounding. Each cycle makes the last one more valuable. Each new model release compounds what they've already shipped instead of making it obsolete.

I see four loops that separate the ones compounding from the ones that aren't.

Loop 1: Product

The best companies I work with have become terrifyingly fast at product validation. PMs are launching Lovable prototypes and getting real customer feedback in days.

When you can validate that quickly, picking the wrong feature just doesn't matter that much. You catch it before your competitor finishes their requirements doc.

Their engineering teams have fully adopted AI coding tools and cut feature time-to-market by more than 50%.

Because they shipped early, they're already sitting on months of real usage data. Hundreds of edge cases. Product quality that's ahead not because of a better LLM, but because the product has been learning in production while others are still in pilot.

They also set up their infrastructure for compounding early on with LLMOps. LLMOps is DevOps for AI Agents. Monitoring, observability, data labelling practices. They have a constant pulse on when their agent makes mistakes. When it does, they can quickly update and benchmark the change. They're not flying blind. They're navigating with confidence.

The companies falling behind here skipped both upstream customer validation, and investing early on in LLMOps.

Loop 2: Operations

While AI product builds represent the biggest upside for vertical software companies, the best aren't stopping there. They're reviewing their entire operations and every function. Customer support. Sales. Marketing. Finance. Same question each time: where can AI reduce manual effort or speed up decisions?

The ones doing this well aren't just deploying tools. They run it like a proper operational transformation, not an IT rollout. They own the change management. They bring their people along. That creates a compounding effect.

Loop 3: Talent

The AI coding tools from six months ago are significantly different from what's available today. Claude is a different animal for office work since Cowork launched. All improving at a pace that requires continuous upskilling to capture the value.

The best companies invest in ongoing enablement across both engineering and non-engineering teams. Not a one-off training. An ongoing programme that keeps people at the frontier.

On attracting AI talent, here's the reality. For most PE-backed software companies it's still brutal. You're competing with big tech and well-funded startups for a very small pool. The smart play is getting your existing engineers to work shoulder to shoulder with external AI experts with the goal of building internal capability. Knowledge transfer compounds.

Loop 4: Portfolio

This is where private equity has a structural advantage that most firms aren't leveraging.

The best PE firms have AI playbooks as a core part of their value creation toolkit. Every pre-existing functional playbook has been updated for the AI-first world. They're treating AI as a platform capability and core IP, not just a portfolio company initiative.

Hg is the best example. Their AI incubator embeds engineers, product managers, and designers directly into portfolio companies. These initiatives already driving new customer bookings up by more than 40% in their first year of AI product launches. And because one company's breakthrough becomes a pattern that gets deployed everywhere else, the whole portfolio accelerates.

That's compounding at the portfolio level.

The impact on hold periods

Hold periods are extending. Multiples are compressing. The selloff added another layer of gravity on top.

What these companies need is escape velocity. Each loop adds speed. Together, they compound into the kind of acceleration that defies the gravitational pull of tougher exit markets.

Hg demonstrated this with GTreasury. They invested in 2023, launched GSmart AI agentic product for finance leaders. Within roughly two and a half years, GTreasury exited to Ripple for over $1 billion. Hg calls it their first AI-driven exit. The AI didn't just improve the company. It made the company attractive to a strategic buyer in a market where exits are hard to come by.

Where to start

You're already doing "AI things". The question is whether any of it is compounding.

Go through each loop and ask one question:

1. Product:are your AI products in production collecting real usage data, or still in pilot?
2. Operations:are operational wins funding the next initiative, or dying as one-offs?
3. Talent:are your teams getting better with AI every quarter, or using the same playbook from six months ago?
4. Portfolio:are learnings crossing between portfolio companies, or is everyone starting from scratch?

The models will keep getting better. The companies that set up their foundations now (proper product validation, LLMOps, change management, knowledge transfer), won't need to predict what's coming. They'll be ready to compound whatever comes next.