Building on a model API or buying a tool: how did you decide?

Summary of public sourcesWritten by our research agent from 5 public sources. Not first-hand experience.

Most teams buy the general pieces and build where the task is specific. Menlo Ventures' 2025 survey of enterprise AI buyers found 76% of AI use cases were purchased, up from 53% the year before. Practitioners describe buying a coding assistant or chat subscription, then writing small scripts or classifiers for their own work. Control over data and spend pulled some toward building.

Leboncoin's story shows both sides. It built an internal chatbot in 2023 so staff could use models without sending data to public services. It later retired it. Demand outgrew a small team, a self-hosted open source option was hard to run, and a vendor's EU data residency removed the original reason to build. For a shared team chat UI, practitioners mostly pointed to existing products. Those who build often put a proxy in front of the model APIs. Proxy docs such as LiteLLM's show budgets by team, user and key, which is often the feature people wanted.

Still open: where the line sits between a useful internal build and a tool nobody maintains.

Sources

• Hacker News: Ask HN: Browser-UI with multi-LLM for the team?
• Hacker News: Ask HN: What is your company doing with AI?
• Survey: 2025: The State of Generative AI in the Enterprise
• Engineering blog: Once Upon a Chat Bot: The Ada Story at leboncoin
• Vendor docs: Budgets, Rate Limits | liteLLM