How do we find our own AI software engineer

Hi all!

We are very happy to announce a big update – a few days ago we finally introduced us AI engineers to the flatlogic platform.

Short story – now you can:

• Work in two environments:
  Dev Environemnt for the development and instant feedback, and a stable environment for persistent changes.
• Control Version:
  Track every change through a system based on our integrated GIT, review the complete history, and roll back change.
• Use AI Engineer for Application Development:
  • Change you data scheme (Entity, field, relationship)
  • Adapt Source code Direct (UI component, logic, force)
  • Update you application data (Permission, role, content)
• Instant Change Reflection:
  All updates carried out through AI are immediately reflected in the preview directly – no need to wait for the transfer.

Now the longer version for those who like details!

We have a very successful launch at Appsumo last year which proves to us Flatlogic Generators are needed!

However, it also shows that it is incomplete-the most requested for the Flatlogic Generator is the ability to make changes to the application produced (clearly-some naive one can think that one generation will be enough :).

Okay, whatever our customers ask – we do! Accepted!

We naively think and promise – we can provide this feature in 6 months. How naive again!

Long story creating our own AI engineers

Okay, so I think for myself – how do we approach “the ability to make changes” from a technical perspective at all? What is the first step?

1. This localizes the problem, of course! Or, in other words, define system limits and degrees of freedom!
   What do we have? We have a web application that is represented as a data model, and we can add entities and fields to the data model, then regenerate applications.
2. That can be done through the scheme editor, but AI cannot use UI – we need to extract commands such as “added entities”, “delete entities”, “added field”, etc., so that they can be triggered programmed.
3. Okay, so AI must be able to carry out this “action”/command on the application – this becomes clear! But how do we do it? We have a chat, but chat is a chat – it’s just a text, not a method of method. So we need to connect AI to our platform and allow it to run orders through several interfaces.
4. Okay, but what about feedback? What if the request for user change requires some command calls? ..
   This one is tough …
   …
5. Then.
   We need.
   To build feedback mechanisms!
   So we need to give feedback on the results of the execution to LLM, for evaluation, and to come with the next step.
   …
   GEEZ! It feels like a solution!
6. So we give a set of commands that can be carried out, feedback on the results of the execution, and ask him to take the next action!
   Cold!
   …
7. Wait, but what about the code? Applications represented as data models are compiled into the source code, the final representation of the software that must also be modified. This one is easy – we add a few more commands: “read_file”, “modify_file”, “replace_file_line”, etc.
8. What if they want to modify the role or permission? Or data in general? Easy!
   More commands: “read_db_schema”, “execute_sql”!
9. Ah, now I get it – we only give all the AI ​​capabilities we have as human operators in our system: Modifying data models, modifying source codes, modifying data. Good! Finally something cool and works!
10. A teammate sent me a guide by Anthropic, who has become classic-https: //www.anthropic.com/engineering/building-efective-agents.
    …
    Mmm, so we also found “AI agent”?
    Good! If people who are respected thinking are the same, then we are on the right track!
11. Okay, the core of the future system is ready! But when AI makes changes, the application must be re -employed, which takes ~ 5 minutes.
    GEEZ.
    Cannot be used completely.
    What are we doing?
    What do human engineers do?
12. They were launched in Dev mode! With hot swaps, or hot module replacement. So this is what we do! We need an environment where the application is launched in Dev mode, so changes are immediately visible. Cold!
13. But what about the state of the application produced? AI implements changes in the application, then what happens? Do we keep it?
    GEEZ.
    …
    I know the answer, but I don’t like it, because that means we have to adjust the system fundamentally. And the answer, as you might expect, is:
    …
    Git!
14. So we arrange our own GIT version of the control system. Even set our own GIT server to host all user applications. So now you just click “Save” and under the new KAP KAP will be made, so whatever you have safely.
15. Cold! Users can now create applications, modify it with AI, and save changes. Good! But all this happens in Dev mode! Not ready for production anymore!
16. So what do we do?
17. Easy -adding other environments that we call “stable” (I hope we will be brave enough to add the production environment as well) where the application is launched in production mode. Now we speak!
18. Ah … now dealing with environmental synchronization. Encourage from one. Reset from the others …
19. A teammate asked:
    – Hey, isn’t everything too complicated? Maybe we need to return to sell a template?
    – NO! We’ve bury the $$$ K, we can’t stop here! This is not what a real man is doing!
20. Now! Everything must be fine, right?
    NO!
    …
    Our pricing model is out of date, never really succeeded, and not transparent. So what do we do?
21. We added a Credit -Based Price Determination Model: Subscribed to get $ x credit per month, spend credit for changes in AI, hosting, and downloading the code of derivative source.
22. Good. Now it feels at least like the first MVP we can publish – the foundation is here.
23. Of course, there are more questions and things we plan to add immediately:
24. Limited Version Free – Yes!
25. More improvements for AI engineers (rags, recipes, memory) – Yes!
26. Add more adjusted business templates, so you don’t start from the basic applications – Yes!
27. More UI diversity – Yes!
28. OPENAI INTEGRATION & STRIPE OUT-OFF-THE-BOX-YA!
29. And many more – https://flatlogic.com/forum/threads/category/roadmap
    (Of course, if we survive, which you can help using and/or spreading news about our products 🙂

Today we released everything until item 18 – it took one year (haha) to get here!

The credit system, which is the final sign before we can say MVP is finished, will be present soon (next week), so you still have a few more days when you have unlimited credit.

One last thing – convergence

It is very interesting that we effectively observe the phenomenon of “convergence” when different actors come with the same solution, do not exchange information from one another: we come to the architecture of “AI Agent” only by trying to solve “the ability to make changes to the application produced with AI problems”, just like many other actors.

The same thing applies to the model context protocol, once again by anthropic. It will be very easy for us to support it because we already support it under the hood – we only need to expose the orders available to our AI agents publicly, so they can be used by other AI agents.