What This Is
An open-source AI coding tool with 200K GitHub stars, OpenCode was completely frozen in CLI mode by a folder left behind during DeepSeek CLI (DSH) installation—not a network issue, not an API issue, not an API key issue, but two star AI tools colliding without any awareness of each other. The developer ultimately needed another AI Agent to spend an hour debugging through 10 leads before locating the root cause.
The mechanism: DSH created a set of npm cache symlinks (junctions—making two directories appear to point to the same location) in the Windows user directory, where the cordis and cordis-plugin-include packages form a circular dependency. When OpenCode launches, it runs a Git snapshot that traverses the current directory; Git doesn't understand Windows junctions and walks into them like a fool, never getting out. That user directory contains over 1 million files, making root-cause analysis extremely difficult.
Industry View
What makes this worth the editorial board's attention is not the bug itself, but the three layers of structural problems it exposes.
First, the AI Agent ecosystem broadly lacks "defensive programming." OpenCode traverses the current directory to take a Git snapshot at startup, but doesn't account for known traps like Windows junction loops. Git, Node.js, and npm all have their own fallback mechanisms; OpenCode doesn't.
Second, the interactions between AI tools are completely unexamined. Two 200K-star projects placed together produce problems—"OpenCode blames DSH, DSH blames Git, Git blames OpenAccess"—a blame-shifting closed loop where no one is responsible and everyone is responsible. This pattern will grow increasingly common during the AI tool explosion.
Third, and what alerts us most: the developer themselves admits they could basically not have solved this problem without AI assistance. Using AI to fix AI-manufactured bugs forms an awkward paradox. There are of course optimistic voices saying this precisely proves the value of Agents—able to call tools across systems and reason autonomously. But the problem is, even debugging the tools themselves depends on other tools. The fragility of this ecosystem runs far deeper than the marketing suggests.
Impact on Regular People
For enterprise IT: When companies trial multiple AI coding or office tools simultaneously, similar conflicts are not unlikely. Before introducing new tools, running a round of "coexistence testing" is more reliable than watching promotional videos.
For individual professionals: When recommended various AI tools, remember they are not "plug-and-play." Many pitfalls you have to discover yourself—vendor demos and real environments are often two different things.
For the consumer market: When some AI product is hyped as god-like, behind it might be just a few open-source components spliced together that blame each other. Lowering expectations of "plug-and-play" gets closer to reality.