This week we were encouraged to start trying to integrate agents into our workflows and see what fits best for our teams. I've used Google for years as a Software Engineer to track down stack overflow, git repositories, reddit posts, etc... . Lately I've been using Gemini more in a conversational(e.g. prompting) format quickly get to the answers I need. This is definitely faster than continuously refining a google search because the AI Model is having a conversation with you. This is just one step from agentic engineering but until this week I didn't really have a reason to fully integrate an AI harness into my daily workflow. The goal of this series will be to document my thoughts as I go down this Agentic Engineering journey because I do believe after working with an Agent that it's a needed tool in my toolbox going forward.
The agent harness I'm currently using is OpenCode . There are many harnesses out there with Claude Code being the one most discussed in mainstream media.
Useful terms I've learned:
- Agent Harness
- Model Provider: Provides access to the models from your Agent Harness
- Agents.md: Is like a README for Agents. It's the first thing an agent will look through when reading your code.
- Agent Skills: Standardized way to give AI agents new capabilities
- Agent Plugins: Collection of skills to form a package
Pros:
- Agents are great for documenting a code base and getting things done that I'd prefer not to do myself or don't yet know how to do.
- The above will help me get my ideas to market faster and ideally help free my time to think new ideas or higher level concepts since I don't have to be in the weeds.
- The barrier to entry for people getting their ideas into the real world has shrunk so more people by using Agents will be empowered to present their ideas to the world
Cons:
- Part of the JOY I've found in programming is solving problems for people and while frustrating to get their at times that sense of euphoria you get when you've found the bug, debugged it for minutes, hours or sometimes even days was part of the fun of the job itself. Agents will take this away as it can solve these problems quicker than you can.
- Eventually I believe some people will allow AI to do all the thinking for them and just blindly follow it which will lead to less useful debate among actual people because there won't be as many people that can cognitively on their own participate in the debate. I still believe it's important to understand at least at a base level the code or whatever product your Agent is putting out because I think you as a individual are responsible for it.
- This will mean less jobs at least the current jobs that are out there. New industries may pop up and maybe training the models and having more custom models for your actual use case will be where the work is but will they pay the same?
Real World Use Case:
P.G. County public schools started this week and unfortunately they didn't have their bus system in a good place. There were at least 10,000 students who weren't assigned buses. You can read all about that here . This is something completely solvable and due to either incompetence or just not enough time this situation was not managed before the school year started. NO KID/PARENT SHOULD HAVE TO HAVE THEIR SCHOOL YEAR START THIS WAY.
AI in my opinion if used right could've freed up the right people to tackle this problem and easily have tested the Chipmunk application that's being used by the school system for load. As a Black man and former Prince George's County resident, I feel we should have done better. P.G. County is majority-Black and led largely by Black representation—it should be a standard for Black excellence, but this falls short.