I find that agents have predispositions about HOW they do things. COLLABORATION with the agents is ESSENTIAL so that implementation is aligned with their predispositions. Think about a human collaborator. Is the goal to be able to provide them an infinitely detailed set of instructions they can implement from the beginning to the end without interruption? Or, do we plan for regular check-ins to discover what unanticipated issues have arisen? This quote about battle can be applied to software development , “No plan of operations extends with any certainty beyond the first encounter with the main enemy forces.” There is no perfect plan. We must have a plan for check-ins and adaptation. We must leverage the strengths of our assets and adaptation around their weaknesses. I find agents are pretty introspective about how they do things. Sometimes I want things done differently. I don’t try to bend them into pretzels to do it exactly like I want to. Implementation plans are a dialog, a synthesis, between what I want and their natural predispositions. Checking in and plan revision is an essential part of the process, not a bug.
2026-05-29 15:40:09
2
Joseph Louis Lagrange :
In my experience, the human/agent split usually hovers around: 80/20 for API design; 50/50 for core data structures; and usually 0/100 for most implementations, followed by human review. If the important algorithms are mathematical, I usually have to rewrite them.
2026-05-30 03:25:26
1
txt0txt :
Because people want perfect control…
2026-05-29 20:04:39
1
notallchads :
oh this hurts. I spent most of the week doing this very thing
2026-05-30 15:51:42
1
JasperD :
This is so true, AI is for augmentation.
2026-05-30 16:00:28
1
Kyngston :
rules are needed only where the model isn’t clear on intent. and all you need to do to find where deterministic rules are needed… ask “whats unclear?”
2026-05-30 00:41:21
1
stidstan :
This is exactly where we are as a dev team, choking the model. I think in part it is the irony of collaborating with an llm on design which has masses of training on traditional engineering and leans in to the engineers fears
2026-05-30 15:20:27
1
Agentic Engineering :
👋🤖👋🤖
2026-05-29 13:36:55
0
To see more videos from user @agenticengineering, please go to the Tikwm
homepage.