You Can’t Grow Feta in Your Garden (Build vs. Buy)
Matt Aliber
Director, Data & AI Product, AdeptID
Think of building a software product like making a great Greek salad. You grow some greens yourself, but there are certain ingredients you’re never going to produce in your own garden. Similar to making a salad, in software the question is no longer whether to build or buy the full solution; it’s which ingredients you should buy.
AI HAS BROKEN THE BUILD-VS-BUY DECISION MODEL
This isn’t breaking news…we’re all living it. AI-assisted development has collapsed the cost and time of building. The variables that the old framework balanced have shifted all at once.
Why build?
| Factor | Description |
|---|---|
| Greenfield cost | Cost to build from scratch is a fraction of what it was. |
| Time to value | Can now beat procuring off-the-shelf. |
| Customizability | Building has always meant easier customization, but with a smaller O&M penalty on every customization as build and maintenance costs drop, there’s less need to procure low/no-code solutions. |
Why buy?
| Factor | Description |
|---|---|
| Technical debt | Legacy code still drags velocity, so buying limits the OPEX required to address technical debt. |
| AI demand | Rising demand for advanced, personalized user experiences is pushing AI into the core product. AI outperforms traditional components with inference or a generative step, exactly where the problem is genuinely hard to solve. |
| Compliance | In high-risk or regulated spaces, compliance is a massive standing investment; often, compliance features are part of the product itself. |
THE NEW DECISION MODEL
Savvy software buyers and organizations are now applying the decision model per-component, not per-product: build the overall product, but buy the high-risk or high-difficulty ingredients. Using the Greek salad analogy again: grow the vegetables and differentiate through design (the proportions, the quality of the ingredients, the dining venue). Don’t raise the sheep, milk the sheep, heat and culture the milk, cut the curd, drain, salt, brine, and age the blocks for the feta. That’s cost-prohibitive and off your core. Outsource the feta to cut risk while keeping your differentiation.
In many modern products, the AI layer is the feta. The software stack now has a new layer, AI inference, whether that’s predictive (a score), generative (a chat), or something else. It’s the feta, not the vegetables: the highest-difficulty, highest-risk, most capital-intensive component. Buy it by default.
Take note of a few longstanding and new feta providers: major corporations that have turned their hardest capability into a composable component that others buy instead of build:
- Stripe reduced payment processing, which is compliance- and infrastructure-heavy, to a few API calls (2011).
- Salesforce launched Salesforce Headless (2026), so you can plug in CRM without taking on the whole platform.
- DoorDash offered a CLI tool (2026) so orders can be placed by AI agents.
WHAT THIS MEANS FOR BUYERS
If you’re assembling a product, apply the following tests when procuring the high-risk, high-difficulty components.
| Test | Description |
|---|---|
| Quality | The feta is a key ingredient, and it had better be excellent. A weak component undermines an otherwise great product. |
| Data | Build your data, not the AI that consumes it. This is the “Bitter Lesson”: general methods that leverage more data and compute reliably beat hand-crafted cleverness over time, so your edge is your domain data, not a rebuilt model. Choose a component that works with your data and extracts more value from it, rather than one that operates standalone. |
| Compliance orientation | Especially in high-risk spaces, do your due diligence on the vendor’s posture. |
| Cost control | Consider SLMs and smart contracting. If you’re using LLMs, insist on LLM agility: the freedom to configure or swap models, or confidence that the vendor isn’t tightly coupled to a single one. |
WHAT THIS MEANS FOR PRODUCT STRATEGY
If you’re building the Greek salad, meaning software for end users, your advantage is in the user experience, your data, and the advantages of scale, not in the components you buy. Assemble aggressively, and differentiate on composition and the experience you build around it.
If you’re building the feta, meaning the component others buy, knowing and investing in your moat matters more than ever. The threat isn’t a competitor; it’s a frontier lab absorbing your capability as a feature. Do the scenario planning: map how a frontier model could eat your lunch, and build toward the positions those scenarios can’t reach. At AdeptID, that moat comes from proprietary data via partnerships, vertical alignment (frontier approaches tailored to the specific needs of the recruiting vertical), and compliance depth in a high-risk space. My colleagues have written about our journey at AdeptID, and we’ll be continuing this series where our technical team digs deeper into the science, compliance, and usage of the AdeptID approach. Read The Demo Gap and Why Frontier Models Won’t Solve Resume Matching.
Differentiation now comes from composition: what you assemble, and the data you feed it, not from building every layer yourself.