Build, buy or have it built: AI agents for a mid-size firm

A 20–500 person firm adopting AI agents has three real paths: build them in-house with internal engineering time, buy off-the-shelf software with AI features bolted on, or have agents built around your own data and processes by a partner. Each path trades cost, control and fit differently, and most firms this size don’t have the in-house engineering capacity that makes the first path work well.

Why “build” is trending, and why it isn’t simple for every firm

Building has gotten easier and more common. In McKinsey’s 2026 State of AI survey, “nearly one-third of respondents (32 percent) report that their organizations have decided against purchasing at least one software product or feature because they were able to build the functionality in-house using agentic coding tools,” and nearly half of AI high performers said the same, against 31% of other respondents (McKinsey, Aug 25, 2026). Note what that measures: skipping at least one purchase, not replacing vendors wholesale. That’s a real shift, but it’s concentrated in organizations with engineering teams built for this kind of continuous, in-house AI development.

It’s also not a shift without failure. Gartner’s June 2025 research predicts over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls — and notes that most agentic AI projects today are early-stage experiments driven more by hype than a clear path to production. For a firm without a dedicated AI engineering function, building carries the same failure modes with less capacity to absorb them.

Buy: off-the-shelf software with AI features

Most practice-management, document and CRM software your firm already uses now has some form of built-in AI. It’s fast to turn on and requires no engineering effort, which makes it the default path for a first step. Its limit is scope: built-in AI generally sees only that one product’s data. It can summarize a document already in that system; it can’t reason across your firm’s procedures, your client history in another system, and the document together, because it was never built to see all three at once.

Have it built: agents built around your firm’s data

The third path sits between the other two: a partner builds agents specific to your firm’s work, running on a data pipeline connected to your actual systems, documents and client information, rather than a generic product’s own database. This gets the fit of a custom build without requiring your firm to staff an ongoing AI engineering function — a partner runs and maintains it as your business and the underlying models change.

How to weigh the three paths

Do you have engineering capacity built for continuous AI development, not just a single project? If not, building in-house risks becoming one of Gartner’s abandoned agentic AI projects: real cost sunk into something that never reaches reliable production.

Does the work you want to automate touch more than one system? If the answer is yes — pulling from your practice-management system, client documents and past decisions together — built-in, single-product AI won’t get there on its own.

How fast do you need the first result, and how much do you expect to add later? Buying is fastest for a narrow, single-system task. Having agents built takes longer up front but each additional agent gets faster once the underlying data pipeline exists, because the foundational work doesn’t have to be redone.

Where Precision AI OS fits

We build the third path: an AI data pipeline connected to your systems, documents and client data, the firm brain that pipeline builds, and agents that work from both — with your people approving anything that goes to a client or that the system isn’t sure about. It’s built for firms that want agents fitted to their real work without standing up an in-house AI engineering team. See the platform, or book a discovery call to talk through where your firm fits.