I built a coding agent.

I built an AI code agent. Why? My goal is to invent an agent that won’t cultivate my laziness to its highest peak.

I have used Cursor, Claude Code, Codex, Gemini, and recently Antigravity, but they are all the same. They are very good at generating code, but I just noticed that I spend more time reviewing what those tools have produced than actually coding myself.

And you know what? I am lazy. The consequence is that I let AI accept generated code without careful review. The moment it works, it is okay for me. Doing this for days on certain client projects, stakeholders are satisfied with the work, but the truth is, when it comes to adding a new feature, I am stuck. I don’t know where to start, or even worse, I don’t know how to design the new feature because I don’t even understand the codebase. I am like a new developer joining a company, asking the AI how my own code is structured.

But now, stakeholders think I can build new features faster. I continue relying on AI tools to do the job for me. Even when I finally decide to check the codebase during the weekend, the technical debt is huge almost impossible to fix. I end up left with two options: spend time trying to fix it, or rebuild the software from zero.

I am speaking for myself, but I am sure many developers or even companies are stuck in that trap. The problem isn't the AI tools; it is us developers. We are lazy, and that won't change.

• We don’t want to craft a better prompt for an AI agent because that means spending more time thinking about the architecture, design systems, and business logic.

• We don’t have time to prepare and give important contextual information to the agent (docs, backend API schemas, etc.).

• We are not taking time to plan tasks; we let the AI agent handle it, we don’t review, we just accept and continue.

What I built is a CLI coding agent(octopus-cli). It is open source, and the developer sets their own API keys. Actually, it supports OpenAI, Anthropic, Gemini, and Grok models, but in the future, it will support local models or even the possibility to add custom model providers.

This is just the beginning. My goal is to build an AI software agent(Platform + Tools) that will consider the laziness of the developer and the lack of responsibility in AI adoption in software engineering.

The agent will have the ability to know every software component (Backend, Frontend, Third-party API definitions) and all documentation related to the software (Architecture, Feature Specs, Figma Designs, and more). With that, the agent will not reside on a single codebase; it will have the capacity to understand the whole project. It will help developers plan new features, review code automatically aligned with technical and business goals, create tasks, help developers with their prompts, and more.

It will be a platform where developers can collaborate on projects easily by leveraging AI capabilities without shipping uncontrolled code.

If you want to learn more or join the adventure, email me at kenmognethimotee@mail.com.