Mid-Sized Companies Build In-House Coding Tools to Protect Data
Commentary on X
Highest ranked
This is the quiet shift nobody’s talking about loudly. Companies want AI capabilities but they’re drawing a hard line at internal data access. And honestly from a security standpoint that’s the right call. The moment your internal MCPs and proprietary data touch an external model you’ve created a risk surface your security team didn’t sign off on.
@GergelyOrosz I wholeheartedly agree you don't want to send your IP. But running a harness locally, and not the model, and declaring victory makes no sense. The model companies are most incentivized to get your IP and the (reasonably effective) models are what's difficult to run locally.
@GergelyOrosz Because internally-built harnesses require incredible effort to stay at parity with what is publicly available for no extra cost. There is ~zero ROI in it. If vendor independence is their goal, they can just use OpenCode, or wrap claude code/codex.
@GergelyOrosz This will happen more often. Once the tool needs access to code, data, tickets, logs, and internal context, it stops feeling like just another SaaS tool. At that point, owning the harness starts to make a lot more sense.
