# Build, buy or partner: a decision guide for AI projects

- **Published:** 2026-04-29
- **Updated:** 2026-10-12
- **Category:** Economics & Pricing
- **Author:** Sinthetix
- **URL:** https://sinthetix.com/blog/ai-build-vs-buy-vs-partner

> A practical framework for deciding whether to build, buy or partner on an AI capability, with the evidence and its limits.

Almost every organization now faces the same question for each AI use case: do we build it, buy it, or bring in a partner? Stanford's 2026 AI Index puts organizational adoption of AI at 88% [2], so the question is no longer whether to use AI but how to source it. The answer differs by use case, and a single company-wide stance usually gets it wrong.

**Key takeaways**

- Decide per use case, not per company: commodity functions lean toward buying, differentiating ones toward building.
- The most-cited evidence favors partnering over building in-house, but it is self-reported, preliminary and should be read with its limits in mind.
- Stable agent frameworks lower the cost of building the plumbing; they do not lower the cost of evaluating, integrating and maintaining a solution.
- Plan for exit: model, data and workflow portability matter more than the initial choice.

## What the evidence says, and does not

A widely quoted 2025 report from MIT's Project NANDA, titled "The GenAI Divide", states that in its sample external partnerships with learning-capable, customized tools reached deployment about 67% of the time, against about 33% for internally built tools [1]. The report itself says these figures reflect self-reported outcomes and may not account for all confounding variables [1].

Treat this carefully. The report describes itself as preliminary findings. Its stated methodology is a review of more than 300 public AI initiatives, 52 structured interviews and survey responses from 153 senior leaders, with success defined as deployment beyond the pilot phase with measurable KPIs [1]. Its own limitations section says the figures are directionally accurate rather than based on official company reporting [1]. Press coverage has quoted the same comparison with different sample counts, so rely on the report. Note also that "partnership" here means a customized, learning-capable tool from an external provider, which is narrower than "buying anything off the shelf". The finding is a prompt for caution about building everything yourself, not a rule.

## What changed in tooling this month

Building got somewhat easier at the infrastructure layer. Microsoft Agent Framework 1.0 reached stable status on April 3 for .NET and Python, with orchestration patterns, checkpointing, middleware hooks and connectors to several model providers [3]. OpenAI's April 15 Agents SDK update added sandbox execution and resumable state; the SDK itself is free, with standard fees for tokens and tools [4].

That reduces the effort of wiring agents together. It does not reduce the work that usually decides outcomes: defining the task, building evaluation sets, integrating with your systems, handling exceptions, and maintaining it as models change.

## The checklist

Score each AI use case against these questions.

**1. Is it differentiating?** If the capability is part of what customers pay you for, or encodes proprietary process, lean toward building or co-building. If it is a standard function (meeting notes, generic support deflection, code completion), lean toward buying.

**2. Does a product already fit?** Be strict. A product that meets 80% of the need with configuration beats a custom build, but only if the missing 20% is not the part that matters.

**3. Where does the data live and who may see it?** If sensitive data is involved, verify retention, residency and training-use settings before buying. This can force a build or a private deployment.

**4. Do you have the people to run it?** Building means ongoing ownership: evaluation, monitoring, model upgrades, security review. If there is no named owner, do not build.

**5. How fast do you need it?** Buying is quickest. A partner can be quick when the problem is well defined. Building is slowest to first value.

**6. How easily can you leave?** Ask for data export, the ability to change the underlying model, and documented integrations. Lock-in is the cost you cannot see at signing.

## Putting it together

| Situation | Lean toward |
| --- | --- |
| Standard function, mature products exist | Buy |
| Core to your competitive position, strong internal engineering | Build |
| Differentiating, but no internal AI or integration capacity | Partner |
| Product fits partially, integration and evaluation are the hard parts | Buy plus partner |
| Sensitive data, no vendor will meet your terms | Build or private deployment |

These are judgment heuristics for framing the discussion; they are not derived from the studies cited above.

## Guardrails whichever route you choose

- Write acceptance criteria and test cases before selecting a route, so every option is measured the same way.
- Run a time-boxed pilot with a go or no-go decision at the end.
- Keep prompts, evaluation sets and workflow definitions in your own repositories, not only inside a vendor's console.
- Assign one accountable owner on the business side.

For businesses, the useful output of this exercise is not a verdict but a short list: which two or three use cases are worth a pilot, by which route, and what result would make you stop.

## Sources

1. [The GenAI Divide: State of AI in Business 2025 (preliminary findings, Project NANDA)](https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf), MIT Project NANDA (PDF copy hosted by MLQ.ai), 2025-07-01
2. [The 2026 AI Index Report](https://hai.stanford.edu/ai-index/2026-ai-index-report), Stanford HAI
3. [Microsoft Agent Framework Version 1.0](https://devblogs.microsoft.com/agent-framework/microsoft-agent-framework-version-1-0/), Microsoft Agent Framework Blog, 2026-04-03
4. [OpenAI upgrades Agents SDK with sandbox execution and a model-native harness](https://www.gigazine.net/gsc_news/en/20260416-openai-agent-sdk), GIGAZINE, 2026-04-16

## How Sinthetix helps

An engineering studio for AI agents, automation, and internal tooling. We build the systems that make operators scale.

[Our services](https://sinthetix.com/services) | [Talk to us](https://sinthetix.com/contact)

Sinthetix is an AI engineering studio. We ship agents, automation, and internal tools for operators who need results, not reports. [About Sinthetix](https://sinthetix.com/about)