I Tried Amazon's Kiro.dev For 5 Days—It's a Below mid-level Dev

I used Kiro.dev 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

How Kiro stacks up vs a senior dev

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 |

\

Final

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 hire Kiro? Definitely yes. We need agents with different angles - just like people - to handle different tasks.