Agentic Patterns

Something like if I do a list of dev pattern and I say:

- caffeinated break for algorithmic thinking improvement

When I'm thinking about an algorithmic logic, go to have a coffee break, and then go back to my work desk to work on it again.

Here is one of the first "pattern" of the project I opened for example:

   Dogfooding with rapid iteration for agent improvement.

   Developing effective AI agents requires understanding real-world usage and quickly identifying areas for improvement. External feedback loops can be slow, and simulated environments may not capture all nuances.

There are so so so many prompts & agentic pattern repositories out there. I'm pretty turned off by this repo flouting the convention of what awsome-* repos are, being the work, rather than linking the good work that's out there. For us to decide with.

Reading this as an avid Codex CLI user, some things make sense and reflect lessons learned along the way. However, the patterns also get stale fast as agents improve and may be counterproductive. One such pattern is context anxiety, which probably reflects a particular model more than a general problem, and is likely an issue that will go away over time.

There are certainly patterns that need to be learned, and relearned over time. Learning the patterns is sort of an anti-pattern, since it is the model that should be trained to alleviate its shortcomings rather than the human. Then again, a successful mindset over the last three years has been to treat models as another form of intelligence, not as human intelligence, by getting to know them and being mindful of their strengths and weaknesses. This is quite a demanding task in terms of communication, reflection, and perspective-taking, and it is understandable that this knowledge is being documented.

But models change over time. The strengths and weaknesses of yesterday’s models are not the same as today’s, and reasoning models have actually removed some capabilities. A simple example is giving a reasoning model with tools the task of inspecting logs. It will most likely grep and parse out smaller sections, and may also refuse an instruction to load the file into context to inspect it. The model then relies on its reasoning (system 2) rather than its intuitive (system 1) thinking.

This means that many of these patterns are temporary, and optimizing for them risks locking human behavior to quirks that may disappear or even reverse as models evolve. YMMV.

https://github.com/EnterpriseQualityCoding/FizzBuzzEnterpris...

For someone looking for a place to start look for patterns where agents do one or more of the following: plan before acting, critique their own work, escalate to humans, remember what they learned, and/or decompose complex work into parallel subtasks. That will cover most of the actual work you are looking for an agent to assist with.

Some patterns I've used frequently enough to codify into skill /agent combos are:

plan then execute

reflection loop / self critique

episodic memory / state externalization

sub agent spawning + checkpoint/resume

CI feedback loop

structured output

context minimization

tool compartmentalization

graceful degradation

Also if you are really looking to dive in get familiar with the Claude Agent SDK. Right now you can use an Oauth token generated from the Max plan even for remote agent deployments as long as you are not serving inference to other users. The Oauth token functionality is NOT mentioned in their documentation, but it works, I'm not sure how long that will be the case.