Nearly every conversation about AI right now includes some version of the same claim: “Because of AI, we can build this ourselves.” It’s a sentiment shared by private equity firms, investment banks, and professional services organizations across the board.
It’s an understandable thesis. AI has made it genuinely easier to build. Developers are more productive, prototypes come together faster, and the barrier to getting something working over a weekend has never been lower.
But working and valuable are not the same thing. The mistake firms keep making is confusing the cost of the initial build with the total cost of staying current. Those two numbers are nowhere near each other — and the gap between them is exactly where the build strategy falls apart.
The car off the lot
One analogy tends to resonate with the dealmakers and partners because it maps to something they intuitively understand: the moment you drive a new car off the lot, it loses roughly 10% of its value. Not because anything broke. Not because you made a bad decision. Simply because the world moved on the moment you committed.
The car is still perfectly functional. It will get you where you need to go. But its market value — what someone else would pay for it — has already declined, and that depreciation continues whether you drive it or not.
Most people accept this as an unavoidable feature of buying a car. What they don’t realize is that the same dynamic applies to custom-built AI infrastructure — except the depreciation is far more severe, and it happens much faster.
Your AI agent has a two-week shelf life
When a firm builds its own AI agent on its own infrastructure, that agent doesn’t depreciate slowly over years. It can depreciate in a matter of weeks. And unlike a car, there’s no trade-in value to recover. The engineering hours, compute spend, and validation cycles that went into building it simply become a cost you can’t recover.
The depreciation doesn’t show up as a line item. It shows up as rework. Across 623 million analyzed code changes, GitClear found that two-week code churn increased 15% while refactoring fell 70% as AI-generated code authorship scaled.1 The result is a development cycle in which more work has to be revisited, rewritten, and maintained after the initial build.
And the pace of AI development makes that problem harder to ignore. New models, frameworks, and platform capabilities are arriving constantly. Each change can require an internal team to evaluate, adapt, rebuild, and revalidate what it has already built. Not once a year. Not once a quarter. Continuously.
The first build is only part of the cost. The real cost is maintaining what you built as everything around it changes.
A car depreciates by 10% the moment you drive it off the lot. The median gap between frontier AI model releases is now just 11 days. The agent your team builds today is already behind the one they could build next week – and the week after that.2
The hidden CapEx problem
This is where the language of capital allocation matters.
When a private equity firm or an investment bank allocates CapEx to building internal AI tooling, they’re not making a one-time investment. They’re funding a perpetual maintenance cycle. Every 11 days, a portion of that investment becomes obsolete. Every 11 days, additional capital has to be invested simply to prevent the system from falling behind.
Firms are built to deploy capital intelligently and to make bets that compound, hold value, and generate returns. They are not built to fund an internal R&D cycle that resets on an 11-day cadence. It’s not a core competency. It’s not even adjacent to one. It’s a distraction that looks like an asset on day one and quietly becomes a liability by week three.
What makes this particularly dangerous is that it’s hard to see in real time. The build feels like progress. The first demo looks impressive. The internal team is proud of what they’ve shipped. But underneath, the clock is already running.
Who should own the risk of innovation?
Here’s the question every firm planning to build should ask: who should carry the weight of a market that isn’t going to stop evolving?
The R&D cycle in AI is relentless. That cost has to live somewhere. The question isn’t whether someone pays it — it’s whether it should be you.
At Intapp, absorbing that cost is a core business function. Our team tracks every model release, every framework update, and every platform shift. We evaluate what changes, rebuild what needs rebuilding, and carry that forward on behalf of every firm on our platform. That investment is distributed across thousands of clients rather than sitting entirely on the balance sheet of one. That scale isn’t just a cost-sharing advantage. It gives us visibility into the best practices and patterns of leading firms across the industry, insight no single internal team could ever build on its own.
For a firm that builds internally, the math is simple and brutal: you are paying the full cost of R&D for something that benefits only you, on a cycle that never ends, in a domain that is not your business. Worse, an agent built entirely in-house only reflects one firm’s own institutional habits. Without an outside benchmark, there’s nothing to stop those habits, including the bad ones, from getting built directly into the infrastructure your team relies on. That risk of innovation is something we need to absorb. It should never fall on the firms we serve.
The ROI case for buy
The ROI conversation around software has always included a buy vs. build component. In most eras, a capable internal team with strong institutional knowledge could make a reasonable case for building. The initial cost might be higher, but long-term control and customization could justify it.
That calculus has shifted. Significantly.
When you buy from an AI-powered company like Intapp, you’re not purchasing a static piece of software. You’re purchasing continuous R&D. You’re purchasing model management, framework updates, and platform evolution as a service. You’re also buying the accumulated learnings from Intapp’s experience across the industry — insights that help you keep pace with peer firms. Every improvement we make to the underlying infrastructure is automatically available to every firm on the platform.
The total cost of ownership for a build strategy is consistently underestimated because teams are rigorous about pricing the initial build and almost entirely blind to the ongoing maintenance cost. When you factor in the engineering hours required to stay current, the organizational drag of managing an internal AI roadmap, and the opportunity cost of the technical talent focused on maintaining infrastructure instead of building higher-value tools like AI agents, the numbers shift dramatically. That shift also means firms may not need to carry as many technical FTEs once platform maintenance moves to the vendor.
This isn’t an argument that buying is always the right answer for every tool in every context. But for the AI workflows at the core of how professional firms operate — those built for specialized, compliance-sensitive, relationship-driven work that defines this industry — the argument for buy has never been stronger.
The strongest argument for buy in a generation
There’s an irony worth sitting with here: the very thing that has made building feel more accessible — the speed and accessibility of modern AI development — is the same thing that makes the build strategy untenable. The faster the field moves, the faster your internal build depreciates. The more capable the new models get, the more quickly yesterday’s implementation looks inadequate.
The agentic era hasn’t weakened the case for buying from a specialized platform. It’s made it stronger than it has ever been, precisely because the pace of change is faster than any single firm’s internal team can sustain.
Across the industry, firms are trying to figure out how to apply AI intelligently — to use it in ways that actually compound, hold value, and serve their clients and their partners in ways that matter. It’s exactly the right instinct, but the path to getting there runs through a platform built to absorb the risk of continuous innovation, not an internal team racing on a treadmill they can’t turn off.
Our R&D team assumes that risk. Not yours.
It’s the only model that makes sense given how fast this space is moving, and it’s the reason the buy vs. build argument — for firms that take the total cost seriously — has never been clearer.
Sources:
1. GitClear. The Maintainability Gap: AI Code Quality in 2026. 2026. Accessed August 21, 2026. https://www.gitclear.com/the_ai_code_quality_maintainability_gap
2. ARK Investment Management LLC, 2026, based on Artificial Analysis data as of April 28, 2026. https://officechai.com/ai/frontier-labs-are-releasing-new-models-faster-than-ever-shows-data/