@mattpocockuk@modemdev@plannotator As an aside, I think people who don't vibe with agentic engineering are kind of here: forever disappointed in the output, not realising you have to work hard at trial and error to get the best out of these tools. A frictionless experience isn't guaranteed.
6/6
Brimming with great takes. Totally agree the real effort should go into team + agent collaboration.
But that is too big brained for me at the moment. So, thought I would share how I protect my solo workflow from markdown fatigue ��
New talk! About the divided human & agent worlds, boundary objects, and building thicker interfaces
I've been exploring how to improve "planning", aka. deciding what agents should build, how you want it built, & making the whole team part of that process
https://t.co/JS73DCdSXi
@mattpocockuk@modemdev@plannotator Then I take those tweaks and friction points back to the plan for a shorter /grill-me. Now i'm back to being "European" and can happily let the agent rip-pppp!!!, because i've already been there. Feels like overkill but it saves so much rework, and my soul dies in rework.
5/6
New talk! About the divided human & agent worlds, boundary objects, and building thicker interfaces
I've been exploring how to improve "planning", aka. deciding what agents should build, how you want it built, & making the whole team part of that process
https://t.co/JS73DCdSXi
@tanishqk Love it. Curious if you've thought about having this sit inside the thing you are developing vs having this pulled out as its own "thing/intermediary"?
@mattrothenberg I think it's great btw. Context is king. Having that sort of "tracing" really plants where a change sits. Like imagine thinking you were dealing with a bug in UI state when in fact it was a server boo-boo.
@mattrothenberg Whoa. I know I'm glossing the main attraction. But, I'm assuming that this is against a monolith codebase? Noticing that the agent has BE context: "What the delete does, on the test database". What was the thought behind that?
@plannotator@modemdev Maybe what I am after is some VS Code extension that allows me to annotate a diff inline. Have that feed back to an agent for live edits. Though weird, seems like everyone else can just reason about a diff "as is". Maybe skill issue.
One difficulty I find with (GREAT) agent friendly code review tools like what @plannotator and @modemdev (Hunk) offers is that I invariably end up cracking open VS Code to see the code "in it's natural habitat". I find it really limiting to see a diff in isolation of the whole.
@plannotator@modemdev My sub-optimal workflow is: checkout branch, soft reset and unstage files, boot up @plannotator guided review, look at the diff in VS Code, use inline annotation tool in plannotator review to instruct agent what I want changed (never use "send feedback" cause keep session).
@plannotator Reviewing large PRs the "traditional" way has been a real pain. Grouping changes so they read like a human wrote them feels like the tool I've been missing. And it's not a cop-out like "get my agent to review your (huge) PR", which I have been seeing a lot of--Ick.
@plannotator Like, I've used Hunk but I find I need more of the surrounding context than just the hunk (I know I can expand surrounding code blocks, I personally find I need file level context). So, it never quite landed.
@mattrothenberg@mattrothenberg How do DS consumers handle "I just need to make this work"? className is the easy path. Remove it and a small tweak becomes: make the case, update the DS, wait for a release. How do you handle this friction? Have people found a way to route around it? "Balance".