Tried Kiro for 5 days - feels like a below mid-level dev
Roman Medvedev
I used Kiro for 5 days to complete my hackathon project (analyzing GitHub repositories).
My quick eval: Kiro feels like a below mid-level dev.
Coding skills: mid-level
Engineering skills: below mid
Discipline: below mid
Here’s how Kiro stacks up vs. a senior dev (based on my 25 years in software):
| What Kiro does | What a senior does | Ideas to fix Kiro |
| Chooses a random repo with 100+ forks | Chooses a few smaller repos | Ask Kiro to research repos suitable for testing |
| Implements one big, long-running command (multiple requests per fork) | Plans smaller steps: show-info, list-forks. Tests step by step → sees most forks are empty and skips them | Ask to force task decomposition (Kiro resists) and break all processes into smaller steps |
| Plans redundant, unnecessary, undisclosed features | Plans only what’s needed | Ask to stay minimal when planning features |
| Rewrites raw data into vague, emotional, emoji-heavy text | Reports raw data as is | Ask to transmit entity names/data directly, without rephrasing |
| Loses insights during planning/implementation | Keeps track of all key details | Summaries + separate notes sessions |
| Starts coding immediately during “specs” discussion | Plans first | Use separate sessions, ask for summaries, and store notes in a separate file |
| Ignores instructions (agent steering rules) | Gets fired | Must follow rules or refund |
| Creates new specs for tiny features instead of extending existing ones | Creates a new package only if reusable | Must respect current session scope |
| Crashes but still marks task as “completed” | Gets fired | Must either finish properly or refund |
| Outputs “successful all done complete” placeholders as results | Raises NotImplementedError | Should always raise for unimplemented features |
| Does a sloppy job | Hunts for a new job | Hopefully more careful with smaller tasks |
| Never runs proper tests | Runs thorough tests | Ask for full test coverage - but beware, your budget may vanish fast |
| Not ready to ship autonomously | Can work autonomously | Add more rules - will it help? |
| Burns through your budget for only uncertain results | Delivers within budget | Pricing should reflect useful results, not wasted usage |
Today both of my 2 access codes were depleted. I’m stuck and have no idea how to complete the project now
My opinion: Kiro isn’t ready to work fully autonomously. It burns through budget fast, delivers only so-so results, and needs tighter rules plus better pricing to be truly useful.
Will I use Kiro? Definitely, but Kiro must first reach a strong middle level. We need agents with different angles—just like people—to handle different tasks.
Originally published at https://hackernoon.com/i-tried-amazons-kirodev-for-5-daysits-a-below-mid-level-dev
Subscribe to my newsletter
Read articles from Roman Medvedev directly inside your inbox. Subscribe to the newsletter, and don't miss out.
Written by