Shipping code is the clearest kind of success: code merged, QA signed off, build light green. We even shipped a functioning AI agent, which feels like extra credit right now. Then users didn’t care. Was it failure, or just the most expensive kind of feedback?

In this session, we'll separate our identity as developers from shipping code and treat “not used” and “didn't help users” as valid, valuable results. We’ll teardown a real deep-research project that moved from an orchestrated workflow to an AI agent, then faced a build-versus-buy comparison against a commercial off-the-shelf tool. For teams using AI, the upside is focusing on proprietary data and integrations, while the cost is walking away from feature work that feels impressive. Even if a feature is canceled, the work can still create value elsewhere.