@tanujDE3180 I do vibecode with Go, but I set up a really strict linting etc to protect a specific style otherwise agent finds a way to make it hard to read but otherwise works great imo
A fun lil prompting trick to get better outputs - instead of writing a prompt, record a Loom / video where you talk through what you want while pointing around your screen.
Then prompt in Confluence "use this Loom to create my spec/vision doc/page"
If you think by talking, it feels much more like explaining an idea to a teammate than writing instructions from scratch. Outputs are often better, too 👌
The best AI development workflow I’ve found uses multiple agents grounded in the same context, challenging one another, with a human guiding the process end to end.
@Atlassian tools enabled me to achieve this. Shared context. Independent judgment. Human direction. Sharing in 🧵
Every engineer has a KTLO / RtB pile they'd rather ignore.
Fix a vuln. Clean a stale feature flag. Update a flaky test. It builds up - and distracts from the actual project work.
So in Jira, engineers are doing something better: Jira workflows / automations assign these tickets straight to agents, which ship PRs end-to-end.
No opening a chat. No CLI. No "hey agent, go do this, do that.”
The workflow just does it. 📷 Here’s one of the examples of how we do it in Jira.
How do you do it?!
#jira #ai #agents #sdlc #rtb #ktlo
🔥 take: your AI isn’t dumb. You’re starving it of context.
A smaller model that knows your codebase, Jira, and past decisions can beat a frontier model flying blind.
Better context fewer re-prompts, fewer tokens, better answers.
What does your AI keep making you repeat?
That’s what changed for me: I stopped pasting context by hand. Rovo now pulls it from my repo + Jira before the agent starts.
The result: fewer re-prompts, less token waste, and better answers.
The model didn’t change. What it knew did.
Agentic workflow at scale is exciting. But scale also means poor context compounds fast.
One under-specified work item = one bad PR. Ten of them triggered automatically = confident garbage at scale.
The difference isn't just the model. It's whether the agent has access to what your team already knows - scattered across docs, meetings and the codebase.
The loop: Enrich → Invoke → Review.
-> The work item is enriched with context from the Teamwork Graph
→ A Jira workflow or automation flow invokes a coding agent (e.g. Github Copilot, Cursor) to draft a PR using the work item
→ You review
What's the biggest context gap your team has hit with coding agents?
@fkadev Yas icin onjyuuyonsai (44) ve yonsai (4), kaca kadar saymayi gosterdi ki buyuk bir sayi versin yas olarak? Henuz bilmedigini varsaydigi seyleri ustune yigmiyor olabilir
@AltugAkgul Her seyin SPA olmasina gerek yok cunku. Ozellikle React camiasi HTTP’nin yillar once cozdugu, tarayicilarin bu cozumleri gayet iyi destekledigi problemleri tekrar tekrar cozmeye calisiyorlar. Sirf look & feel veya SSR’dan kacmak icin Amerikayi tekrar kesfetmeye gerek yok