2 Comments
User's avatar
UtsavChokshi's avatar

Good read.

Please add official website for deep agent to help readers :

https://deepagent.abacus.ai/

Context section is helpful.

Few questions :

1. What does implementation completeness mean ? It would be awesome if you can define that for Scenario 3.

2. What do you mean by extracting deep agent's patterns ? I have faint idea. It would be good if you can define.

3. It would be great if you can add your own nuanced engineering perspective on final results produced ?

"""

I spent hours with Claude Code, Cursor, and my own engineering eyes evaluating which structure is best and which implementation is the most robust.

"""

4. [noob question] One of biggeste takeaway for me is : "Deepagent creates compliant app". Are we going by their claims or there is a standard way to check compliance ?

Are you suggesting implementing something like audit logging makes it compliant ? Would you consider that as pattern that can be extracted out ?

Expand full comment
Stanislav Huseletov's avatar

Fantastic questions, and glad you enjoyed the article.

1. In Scenario 3 DeepAgent produced full working app with a database integrated and an interface to interact with the database as well as functioning backend and frontend. Claude (in 1h) generated a very strong architecturally backend with prometheus monitoring and Docker-ready deployment, but the frontend was lacking, and of course there was no completely integrated DB unlike DeepAgent

2. I meant to use the things that worked well for them, like multi-agent usage, iterative inner-agent workflow, no going-back function, and a turnkey deployment with a focus on database compatibility.

3. As I mentioned above, its pretty impressive what all of the tools achieved in my purposefully-time-limited 1h interval I gave to each. I left all the results in the repo attached, but ultimately my engineering take is in how I evaluated the scalability of the apps build.

I made a point to look at architecture, data management, cache management, authentication implementation (where relevant), readability, documentation, UX, and GUI provided by each of the workflows with my engineer-eyes, the way I would look at someone's project handed off to me to improve further.

4. Here I admit I did not do any testing and just took their word for it as I doubted they would lie and, more importantly, I have no expertise in this area to really validate anything beyond simple research of the very limited secondary sources.

Expand full comment