Day 10: I'm getting a bit smarter in how I think about agentic systems and LLMs especially. I was playing around with the idea of creating more professional looking starter templates for the system to choose from.
Then I remembered I sucked at design so started looking for designs. But then I'd need to recreate the designs as templates.
Then I thought why not feed the designs to the platform and have it make the starter templates (still might do something like this). But then I thought, for now, why not let the user upload their own design. So I added this feature and...
#buildinpublic #AgenticAI #AWS #OpenAI
Day 9: The video is a full build based off a user request of "create an app that helps me track my meals and calories, it should let me log what I eat, keep a daliy count of calories consumed and let me view past meals. populate with mock data and images as we will add a backend later"
#AgenticAutomation #NoCode #buildinginpublc
Day 8: focused on the Dev agent this morning because I want to build more complex structures. After pushing I asked to build a X competitor and provided a little more detail than normal.
The results were overall pretty good, except the QA agent had to run a few times to fix issues.
I plan to look into what happened and get the Dev agent commits to build on the first try again. but I like the outcome
#BuildInPublic #buildinginpublic #AIAgent #NoCode
@Daily_BiP
Day 7: Missed a few days but made some great progress. Turning my focus fully to feature updates and some app design enhancements. Right now the Dev agent is consistent at creating apps on the first pass with the occasional assist from the QA agent.
Also, will start to think about opening up an alpha
#buildinpublic #buildinginpublic #OpenAI #NoCode
I've been building this AI mobile app generation platform using flutter and backed with OpenaI LLMs (v0 for mobile I guess), mostly to understand the current capabilities of LLMs and how I can take advantage of it. Curious to get honest feedback on the progress and the concept.
I hope to open it up one day as a beta if the interest is there and I make enough progress. I just started my #BuildInPublic journey to keep myself accountable. Prior there were too many days without me touching it.
Today I built a workout tracker. This also creates the iOS and Android versions behind the scenes. If there's interest, I'd love to get feedback on the roadmap as well
Day 6: Busy making tweeks to the Dev agent so it can consistently build the first pass of the app without the QA agent needing to fix anything. Also trying to ensure the other agents don't overdue instructions that cause the Dev agent to build too much.
Getting there.
Once this phase is consistent, I'll likely move to follow-up requests (features) from the users.
Not sure if I will need to incorporate something more sophisticated at this phase. I may beging testing with some agents backed by Anthropic LLMs on Bedrock.
So far OpenAI is fine.
#CodeWithAI #OpenAI #AWS
βThere will be an AI software engineer by the end of the year.β
@jillchase124, partner at CapitalG, breaks down how AI companies need to stay ahead of this reality, at our Sessions: AI event.
Day 5 (Part 3): This build the QA agent kicked in again but had some issues that required some fixes and updates. I triggered the build multiple times to get the failure so the agent could respond. Finally it ran successfully and I think I hardened QA...a bit.
This agent doesn't respect the requirements, so I have to look into that. It especially destroys the theme as it fixes issues.
I have an end state in mind for this platform, but it seems far away - although I have to say I'm still pretty pleased.
#Flutter #CodeWithAI #BuildInPublic
Day 5 (Part 2): This was a cool build at first in that the QA Agent kicked in after the build failure and fixed the issue to kick off a new successful build.
Problem was, the logic. The Dev agent pulled one of these:
// TODO: Implement real move logic. Currently always true.
return true;
In any case, it's a victory. More prompt tuning and validation to come
#CodeWithAI #BuildInPublic #AWS #OpenAI
Day 5: Connected the frontend to the platform. I can now select apps (conversations) and continue from where I left off.
The video is the task app the dev agent built in one pass. I'm satisfied at that accomplishment but obviously many issues.
First, the look. There are many options to manage this better, so not overly concerned.
Second, the messages from the agents need to be more...conversational and understandable.
Third, the web preview quality is degraded compared to the iOS Simulator. Not sure how much that will matter but I feel like it will.
Fourth, follow-up requests - lets call them features - won't be trivial as far as I can tell. At least not in the current architecture but I'm not ready to delve into anything more complex until I nail the first pass.
#buildinpublic #nocode #mobiledev
Day 4: Success!
But not really day 4, just 4th day of posting, but who cares. Anyway, the dev agent built a working task list app. Looked terrible. Really awful but I think it fulfilled the basic requirements. Need to check later, but..SUCCESS!
#BuildInPublic#AItools
Day 3: just realized I should lead this way - although I started this dev effort before I began posting about it. Anyway, trying to get my dev-agent to build an app in a single pass. nothing fancy just a task app or something similar for now. It's one step forward, two steps back.
Getting close. Just reading the build errors and adjusting the prompt or the metadata and trying different OpenAI models. #BuildInPublic #OpenAI