← sources

Feeling like AI is a thekedaar from construction world.

source: reddit (reddit:developersIndia)published 2026-11-13 09:34ingested 2026-11-14 02:36

retained text

So my company is heavily invested in AI. We built an end to end automated system to resolve bugs and deliver small features.

1. Product Team reports bugs on JIRA 

2. AI agent checks the details and tries to reproduce in an dev environment. On success it lists down the steps and moves ticket to Accepted.

3. Accepted tickets are picked up by a cloud cursor agent, which creates a PR to fix the bug or implement the feature. 

4. PR is reviewed by 2 AI agents -> First one checks the code quality (and static checks like sonarqube, audit reports, linting, formating etc) and second one checks the architecture and company defined standards. If both passes then pre-commit is triggered. 

5. After successful pre-commit (unit tests, vulnerability report, code coverage tests), another AI agents checks for merge conditions, classifies PR as low, medium, high or critical. Low PRs get auto-merged to bugfix / feature branches, others require human approvals before merge. 

6. Once merged Post commit pipeline triggers and generates a docker artifact for the service. Automated pipeline deploys it to dev environment and triggers a api test and integration test suite. If both passes then code is merged to main branch and a stable tag is created and JIRA ticket is move to pending validation state. 

7. Another AI agent picks up the ticket and uses the steps to reproduce the bug created as part of \[2\]. If the bug is removed then the ticket is moved to pending sanity state if not additional details are added to the ticket and moved back to Accepted state. For User Stories the QA team needs to define the acceptance criteria, essentially an markdown doc attached to the jira ticket. 

8. Pending release tickets are deployed in sequence using automation pipeline to QA env for regression testing, PRE-PROD for chaos and load testing and STAGING for sanity suites. If all of them pass ticket is moved to pending release state. 

9. An automation pipeline picks up tickets in pending release state every morning 10AM IST and deploys them to production in canary mode. The reporters and the service owner receives email notification once completed. 

10. All the user base is divided into groups like internal, plan-a, plan-b etc. PMs have access to a UI through which they can move users to a canary version of a particular feature or a particular bugfix version. 

Last week we delivered on one of the features using this pipeline end to end. Took a total of 44 hours from creating the user story to production deployment. 

Now while this sounds all exciting, the quality has gone down significantly. We are shipping things faster but started to get more bugs reported in turn. Our pager system is buzzing all day around, somedays even getting 100s of incidents being reported. 

Feeling like AI has become thekedaar from construction world, sure we get a usable product fast and cheap but the quality is not up to mark and you have no idea what you'll run into. 

document fetched 2026-11-14 00:24 · canonical https://reddit.com/r/developersIndia/comments/1xv3jqf/feeling_like_ai_is_a_thekedaar_from_construction