Instructor library creator Jason Liu requests feedback contrasting micro and macro workflows for AI coding agents
Macro spans persistent automations; micro focuses on iterative steering.
Users praise GPT 5.5 low as surprisingly good when discussing micro versus macro examples for coding agents.
No Digg Deeper questions have been answered for this story yet.
Most Activity
@jxnlco micro = low reasoning macro = xhigh
micro = well-defined task (sometimes even large ones) macro = discovery, taste, refinement, reflection
Macro - agents .md - docs the agent writes and reads on demand, aka progressive disclosure system for an extremely large “system prompt” - automations - timed clean up sweeps - bug bot & automated reviews - analytics & bug reports triggering flows - plan .md + goal - skills
Micro - what you type as prompt - how you guide the agent and steer it or course correct - how you review output & make changes and edits - The process of doing the work, discovering what you want, and what feels good. Taste through trial and exploration
i want to write a blog post on codex 'micro vs macro'
what examples do you have about how you micro and macro coding agents
@jxnlco Micro: validate this bug.
Macro: orchestrate the entire process of doing recon, js analysis, exploring the app, finding vulnerabilities and primitives for vulns, chain multiple primitives together, submit the findings
@jxnlco micro: give the agent a broom and one dirty room.
macro: give it a map, budget, constraints, and let it discover which rooms are fake.
most people fail because they ask for macro outcomes with micro context.
@jxnlco Cool. In practice, the micro/macro distinction for me has become more about how much reasoning power needs to be applied, and less about the task size. How much of the outcome is understood and well-explained vs parachuting codex into a complex project.
@jxnlco what would symphony be on this scale? is it a macro orchestrator for micro agents?
@JoshHimwich @jxnlco GPT 5.5 low is surprisingly good
@jxnlco Micro is "fix this failing parser test, stop at the diff."
Macro is "ship Slack import end to end, touch schema/UI/docs/tests, hand back the PR and the risks."
The boundary is where I stop steering every file.
I see micro and macro coding agents as different levels of focus in execution.
Micro agents are best for clear, well-defined tasks, such as fixing a specific issue or adding tests for a single API path. Their goals are straightforward with tangible results, code and proof of functionality.
Macro agents, on the other hand, are used when we need to think broadly about the bigger picture. They help determine priorities, plan implementation, and break broader strategies into actionable steps.
Macro agents help define what we should aim for next, while micro agents help achieve those goals in specific tasks.
However, using macro agents without sufficient context can lead to vague plans, while using micro agents without clear boundaries can result in excessive rework. The best approach is to use them together: macro agents for strategy and planning and micro agents for focused implementation, testing and clear outcomes.
@jxnlco micro = turn by turn steering macro = /goal
@jxnlco been running dual agents lately - one handles the 10 line refactors while the other maps out entire feature branches
the micro one catches edge cases in real time, macro one prevents me from architecting myself into a corner
what's your current split looking like?
@jxnlco bug validators(micro), autonomous (/goal macro)
@jxnlco micro: agent changes one function, runs tests, checks lint. whole context fits in one read. failure is cheap. macro: agent restructures a module, touches 12 files. needs checkpoints because the diff outgrows the context window. failure means rollback not retry.
@jxnlco Здравствуйте. Скажите пожалуйста, а есть ли передача данных в режиме реального времени, результаты симуляций, анализа, отчёты от агентов помошников агенту, который ведёт всю линейку проекта? У нас все микро и Кодекс предпочитает держать руку на пульсе, мы задачи не делегируем✨