Key Takeaways

  • OneGuide's State of Portfolio Ops Report reveals that 51% of private equity firms report software engineering efficiency gains from AI, but fewer than 6% achieve a 10% EBITDA lift.
  • Executive interest in democratizing development allows non-technical staff to ship AI-generated code directly to production, bypassing standard engineering filters.
  • Pushing unvetted code into production violates Little's Law; increasing work-in-progress inventory clogs the system and degrades overall throughput.
  • Dave Mangot argues that deployment frequency, rather than AI prototyping speed, remains the accurate test of an organization's software readiness.

The 20-Minute Production Trap

Engineering leadership faces growing pressure from non-technical executives who believe software creation should no longer require software engineers. Portfolio executives see generative tools as an end-run around development backlogs. Dave Mangot pointed out that this pressure often targets the wrong part of the delivery cycle.

“And like the thing that I'm struggling with the most right now when I talk to CTO's is this idea of we're going to democratize development,” (play from 15:49) Mangot said. “Anyone in the company can come in on any day, sit down with some AI agent, talk about the thing that they want in the product, and in 20 minutes, the agents will have written a whole new implementation and it'll be shipped out to production.” (play from 16:03)

This setup ignores the operational reality of enterprise software. Writing code is rarely the primary constraint; maintaining architecture, managing dependencies, and protecting system stability take up the balance of technical labor. When non-engineers generate features on demand, they bypass product discipline.

“So, like if the receptionist decides that our accounting software needs some kind of special thing, they he should be able to come in, sit down with an AI agent, and 20 minutes later that thing that the receptionist thought of should be in production,” (play from 16:25) Mangot noted. “And I'm like, I hear what you want to do, but I'm not really following why that's a good idea.” (play from 16:42)

Little's Law and Pipeline Bottlenecks

Data from OneGuide's State of Portfolio Ops Report exposes the split between technical activity and financial performance: 51% of private equity firms see software engineering efficiency gains from AI, but fewer than 6% see a 10% EBITDA lift. The productivity gains vanish before they hit the income statement.

The Private Equity Funcast hosts explained that this breakdown matches classic industrial theory. “It's a little bit of I always quote Little's Law, the manufacturing law that having more WIP doesn't get more through the factory. In fact, having less WIP gets it through the factory.” (play from 20:46)

When teams use AI agents to flood repositories with new code, they overload manual QA, automated testing suites, and data infrastructure. The work-in-progress piles up immediately in front of downstream constraints. Adding code volume to a system with slow release pipelines only lengthens lead times and increases defect rates.

Solving Customer Demand Instead of Prototypes

Speeding up feature output also creates an illusion of progress inside product teams. An AI agent can spin up an interactive mockup or functional feature in an afternoon. That speed often tempts product managers to behave like assembly-line workers rather than commercial evaluators.

“I don't really want my product people spending a lot of time on prototypes and and that's probably what AI is good for,” (play from 19:14) Mangot said. “But what I do want them to do is to understand what my customers want.” (play from 19:22)

Accelerating bad ideas into production costs capital twice: first to build the unvetted code, and second to maintain or remove it when users reject it. Engineering readiness cannot be measured by how fast an agent drafts pull requests. It is measured by deployment frequency, code quality, and whether the finished output solves an actual customer demand.

Why It Matters

Investment theses that model immediate software margin expansion via automated code generation misjudge where software costs live. Flooding production with unvetted code creates architectural debt that slows release cadences and inflates maintenance costs across a five-year hold. Capital allocation models that price in automated development will struggle until portfolio companies resolve downstream QA and operational bottlenecks.