Every time an agent makes a wrong turn or pulls you into the loop unnecessarily, ask yourself: how do I prevent this from happening again? It could be a new rule in a steering file, a skill that captures a procedure it got wrong, a command line tool that automates a manual step, an MCP server to gather the right context, or wider access to a tool or system you had previously kept out of its reach. Each of the agent’s mistakes and interruptions is an opportunity to extend its autonomy further.
This is the new daily habit of frontier engineering: you’re not just building the product, you’re constantly improving the system that builds the product.
Over time, this habit compounds. You start with agents that only run when you explicitly prompt them and require your input often. As your steering and tooling matures, agents become more autonomous and can run in the background: ambiently finding and fixing bugs, cleaning up tech debt, and improving code quality without you initiating a task.
As models improve, newer agents can clean up after what older models generated. When a new model comes out, re-evaluate whether workarounds you built for the old model’s weaknesses are still necessary. Your steering files and tools are living artifacts that evolve as fast as the models do.
Every time an agent makes a wrong turn or pulls you into the loop unnecessarily, ask yourself: how do I prevent this from happening again? It could be a new rule in a steering file, a skill that captures a procedure it got wrong, a command line tool that automates a manual step, an MCP server to gather the right context, or wider access to a tool or system you had previously kept out of its reach. Each of the agent’s mistakes and interruptions is an opportunity to extend its autonomy further.
This is the new daily habit of frontier engineering: you’re not just building the product, you’re constantly improving the system that builds the product.
Over time, this habit compounds. You start with agents that only run when you explicitly prompt them and require your input often. As your steering and tooling matures, agents become more autonomous and can run in the background: ambiently finding and fixing bugs, cleaning up tech debt, and improving code quality without you initiating a task.
As models improve, newer agents can clean up after what older models generated. When a new model comes out, re-evaluate whether workarounds you built for the old model’s weaknesses are still necessary. Your steering files and tools are living artifacts that evolve as fast as the models do.