The MyrmexIQ blog
Field notes on AI search, engineering analytics, RAG, and coding agents — what worked, what didn't, and why.
-
From Jira ticket to pull request: living with an AI coding agent for three months
We ran an AI coding agent that breaks down tickets, writes code, and opens PRs against our own repos. Here is the honest split of what it does well and what we still do by hand.
Read → -
Per-tenant encryption and data isolation: how we keep customer data separate
A plain-language walkthrough of how MyrmexIQ isolates each customer's data and why every tenant's integration credentials are encrypted with a different key.
Read → -
Investigating an incident across five tools without opening five tabs
When the error rate spikes, the answer is scattered across logs, monitors, the last deploy, and a Slack thread. Here is how our cross-source investigation tool assembles that into one timeline — and where it still needs a human.
Read → -
Cutting engineer onboarding from weeks to days by making tribal knowledge searchable
New engineers spend their first weeks asking questions that have already been answered somewhere. We measured where the time goes and what actually shortened it.
Read → -
Auto-generating the weekly status report (and why the first ten were useless)
We tried to have AI write our status reports from Jira data. The early drafts were confidently wrong. Here is what we changed to make them usable.
Read → -
RAG on real Jira data: the chunking decisions that actually mattered
Retrieval-augmented generation demos look easy. Pointing one at ten thousand real Jira tickets is where the interesting problems start. Here are the ones that changed our results.
Read → -
The decisions that only happened out loud: making meetings searchable
A surprising share of engineering decisions are never written down — they happen in a call and evaporate. Here is what changed when we started indexing meeting transcripts alongside tickets and Slack.
Read → -
Catching the same bug twice: a practical look at bug pattern detection
Recurring bugs are expensive because nobody realises they are recurring. Here is how we group similar issues over time and what the patterns actually tell you.
Read → -
Keyword search vs. agentic search: a concrete example from our backlog
"Agentic search" gets used loosely. Here is a real question run through both a keyword search and an agent that plans, searches, and correlates — with the actual steps.
Read → -
What we learned putting Jira, GitHub, and Slack behind one search box
We spent six months routing engineering questions through a single AI search instead of four browser tabs. Here is what actually changed, and what still does not work.
Read →