Why OpenCode became my coding workhorse
I briefly experimented with Claude Code, then began using OpenCode as my main coding agent. That is my experience, not a claim that one tool is best for every developer. What mattered to me was having an agent work in my development environment while I stayed responsible for its changes.
The shift was practical: I could hand off a bounded task, let the agent inspect and modify the project, then review the result. I still decide what to build, whether the change is sound, and what to keep. That workflow is why OpenCode became useful to me; it does not make the agent an authority.
The Hacker News discussion of OpenCode offers a separate set of user reports and criticism. I distinguish those comments from my own experience and from documented product behavior below.
The workflow that worked for me
I get the most value from tasks with a clear outcome and a scope I can inspect. I describe the expected behavior and relevant constraints, then review the diff and run the checks appropriate to the change. If the result is surprising, I investigate or revise the task rather than assuming plausible-looking code is correct.
That is my personal workflow, not a special property of OpenCode. It reflects the way I prefer to use coding agents: delegate a specific piece of work, then verify it myself.
What I value about OpenCode
OpenCode’s open-source nature was part of its appeal after my brief Claude Code experiments. It gave me a project I could explore and configure. Open-source availability alone does not establish that a particular setup is reliable, private, or secure; those questions depend on the software, configuration, provider, and environment.
OpenCode documents configuration options and project instructions in its configuration and rules documentation.
Configuration, models, and reported defaults
Configuration matters because it can affect which provider and model OpenCode uses and how instructions are supplied. OpenCode’s documentation covers model selection and provider configuration. Check the documentation for the version you install rather than relying on older recollections. (Models; configuration)
In the Hacker News thread, commenters reported different understandings and observations about model defaults and lightweight tasks such as session-title generation. Some said they believed requests went to a different service or model than expected; others disputed those interpretations or referred to documentation and subsequent changes. These are commenter reports from that discussion, not verified statements about current product behavior. I have not independently tested the claims.
When the provider or model matters, consult the current documentation and explicitly check the selection in your own setup. Treat claims about data handling as specific to the version, configuration, and provider involved—not as settled by a comment thread or by OpenCode’s open-source status.
Permissions and network access
OpenCode’s documentation describes permission controls for actions available to the agent. Consult the permissions documentation for the options and behavior applicable to your installed version.
In the Hacker News discussion, commenters raised concerns about permissive behavior, configuration precedence, and network access, and suggested restrictions such as limiting internet access in an execution environment. Those are commenter reports and proposals, not proof that every installation behaves alike or that any one setup is universally safe. I have not independently verified those reports.
For my own work, I would match an agent’s access to the task, keep credentials out of reach where feasible, and review consequential commands and edits. Isolation can reduce exposure, but it is not a substitute for understanding what the environment can access. Decide whether network access is appropriate for the project, and verify the behavior of the specific setup you use.
Release cadence and stability
In the Hacker News thread, some participants described OpenCode’s release cadence as frequent and disruptive; others reported concerns about bugs, complexity, or resource use. Those are commenter experiences and opinions, not measurements I can generalize to every user or release, and I have not independently verified them.
For a workflow that depends on stable behavior, check the project’s release information and test updates against a representative task before relying on them broadly. OpenCode’s releases are listed on GitHub.
What I would tell someone trying an agent
Start with a task whose success criteria you can check. Review the changes and run the relevant tests yourself. Learn the model and permission settings before putting confidential code or data in scope, and verify behaviors that matter on the version you have installed.
My path—from brief Claude Code experiments to using OpenCode as a workhorse, is one person’s experience. It explains why I value an open-source option and why I treat an agent as a collaborator rather than an authority. The Hacker News discussion adds user perspectives, but those reports do not replace product documentation or testing your own setup.
Sources
- OpenCode configuration
- OpenCode rules and project instructions
- OpenCode models
- OpenCode permissions
- OpenCode releases
- Hacker News discussion of OpenCode, commenter reports and opinions referenced in the article; not verified product documentation or evidence of current behavior.
opencode became my coding workhorse