ake2l

Coding Agents make building systems incredible fast.

I always say "The first prompt is the cheapest and look like magic", but what comes after is engineering work.

After 100 prompts, do you still understand what the agent built? Is it still the architecture you wanted?

Thats the reason I am building ArchKeel.

I define the target architecture with the agent.Components, responsibilities, dependencies and constraints.

Both of us can explore the same architecture

As-Is shows what is actually implemented ( whatever is messureable ) Target shows what we agreed on Diff shows where we still have a gap between As-Is and Target

So the agent and I have a common ground to discuss, challenge and fix things.

Open source. Python, TypeScript, Dart.

https://github.com/rapiddweller/archkeel

How do you make sure your agents are still building the system you actually want? And how you get a deterministic trustable reflection of what is implemented ?

For me the interesting part is not whether the model gets better. It probably will. The problem is when the same system creates the change, explains why the change is correct, and basically also grades itself. “Just review it carefully” does not scale either. After 50 correct looking changes humans start trusting the green output. I do too. I am experimenting with moving more of this outside the agent ... deterministic checks, frozen behaviour, architecture constraints, explicit evidence. And probably most important -> UNKNOWN when I simply cannot prove something. I increasingly think this is the missing layer in serious agentic engineering. Not another smarter agent judging the first agent, but boring independent machinery which does not care how convincing the explanation sounds.

I think it is not needed to own every line of code anymore , but you should still have control and idea about the architetcure of the system. You should understand what componenets are there and what is the repsonsibility , how are the orchestrated and what are my quality gates where i messure if what i requested match the results. Thats why i build archkeel (https://github.com/rapiddweller/archkeel) and datamimic (https://github.com/rapiddweller/datamimic) ... to increaes the transparency and review surface for human ... to make the results easier to judge ... i think when we stop knowing anything, we can also stop burning token and ressources

Yes. DATAMIMIC CE can model relational constraints including foreign keys, composite keys, parent/child relationships, cardinalities and uniqueness, and generate the corresponding data across the supported relational databases. The authoring api is already configured the way it is guiding the agent to model the existing relationship;ationship right. There is always pace for improvement and automate more. In our EE Platform we have a separate Service what is dealing with project environments, Schema scanning, recommendation pipelines ( for PII scanning and recommending Converters, Generators ) and build a DATAMIMIC model based on this. In the CE we only have the parsing and execution engine of the models.

I’ll change that. Thanks for calling it out.

:)

We've just released DATAMIMIC Community Edition under the MIT license. It's a Python-based platform for generating realistic test data using a model-driven approach, originally developed for enterprise clients who needed to handle complex data structures while maintaining GDPR compliance. Core features:

Model-driven architecture using XML configs Built-in data anonymization and pseudonymization Support for JSON/XML/CSV/RDBMS/MongoDB High-performance Python core (3.10+) Extensible through custom generators