Hacker News new | past | comments | ask | show | jobs | submit login
Show HN: AI-driven workflow manager built for developers (vairflow.com)
3 points by csamal 7 months ago | hide | past | favorite | 2 comments
We are building a AI-driven workflow manager that's built for developers and enables your team to stay lean, build faster and reduce cost. We are collecting feedback on the idea and will launch in a month. Feel free to share your feedback.

In the example where you specify removal of user logs. How does "AI" figure out user logs are in mongodb?

Our first iteration of product we built was a workflow engine similar to what you built. It is hard sell. You are competing with developers who can script this efficiently and faster than learning a new framework. In a scenario AI generates wrong script, it can lead to liability. Also market is fairly small for the workflows you specified in the example. Market is much larger when it comes to business process automation which is quite saturated right now.

When user logs in, they can integrate their Github repo and add / integrate their resources such as MongoDb or any database. From the database, we can get the schema of the data and AI can understand how it's used in the codebase. So, when we say in natural language "Delete user logs....", then from schema and usage in codebase, we can link to a collection say "user_log". That's why we have a "Resources affected" section where developers / reviewers can see if what our AI predicted is true and edit it. This feedback loop improves our AI's next prediction.

What we shown in landing page is a much smaller subset what our vision is. We don't want to just generate code that looks good to developers on first try. We want to generate code that will pass all test cases in backend. Test cases here are not limited to unit tests, it could be FE browser based automated tests through Selenium, Mobile app emulator test cases. So, our AI has feedback loop on what click in Frontend, Mobile app generated any error and go back, regenerate code that fixes the error automatically.

AI script generation is not final in our case. Developers and Reviewers have final say and edit them. There's no issue with liability here. As the process is same but X times faster. Someone is reviewing in the end, before it's merged to codebase.

It is true that developers can script it efficiently. But the cost of hiring, vetting full-time developers or even freelancers is not negligible. We will integrate it with clients and freelancers after our AI and the workflow matures. So, for any client, the task which took 100 hrs and $5000 to develop by a full-time developer or freelancer working sequentially, will take 10-20 hours (leveraging the parallelism in the workflow), partly done by AI and rest done by freelancers all over world, with total cost say $500-$1000.

Whether we do it or someone else, I am definitely confident of a future where there will be no full-time developers. There will be only freelancers. Most of the work will be done by AI and rest will be done by freelancers all over the world, without client ever vetting or even seeing the freelancers. Like "Uber for Tech team", which scales up and down as per tasks.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact