Ah! So artificial samscrolling is still just scifi! "I canned the first one because the AI wandered off building a whole new text and frame buffer setup. A bit more guidance led now to a version where "
Yes AI is basically a micromanagement course. You have to crush all simulated attempts of self initiative. Learning this well is how you make sure you never work with motivated humans ever again. Thx for sharing the struggles, and the warning. On Wednesday, January 28, 2026, Paul Lalonde <[email protected]> wrote: > You're absolutely right, Hiro. I neglected to point at the code, in a > side branch in my edwood fork, largely out of embarrassment for lack of > features. > > You can get it now at https://github.com/paul- > lalonde/edwood/tree/Markdown-take-two > > The code assets have effectively zero value - they are entirely slop, and > I have expressly not read the artifacts, and though the tool is useful > there are a significant number of code organization failures. > You'll find the "design docs", which are half-slop, part-human markup and > iterated, in the docs/ folder. They are haphazard and of various levels of > quality. There's almost certainly not enough there to make building the > markdown functionality "hands-free" and reproducible. > > This is my second attempt. I canned the first one because the AI wandered > off building a whole new text and frame buffer setup. A bit more guidance > led now to a version where the win has a text with two renderers, and a > mapping layer for selection positions from one to the other. > Sadly, it wandered off into writing its own interaction routines and so > the scroll bars (check out the horizontal scroll bars for code blocks and > other un-wrappable structures) are not integrated, and > dragging/cutting/pasting is mostly custom, though calling the underlying > model on the text. > > It's messy. And a ways from doing wysiwyg editing of the markdown, but > not as far as I would have expected in the 8-10 hours of work this has > taken. That's after stalling at doing basic rich text at least 3 times in > the last 20 years. > > Paul > > > On Tue, Jan 27, 2026 at 7:50 AM hiro <[email protected]> wrote: > >> Did i miss a link to your git repo with the prompt/context/markdown >> files? Or is the prompt not ready to share. >> >> I agree the idea expressed in the prompt is often more valuable than the >> code output itself. Human Programmers often keep that part private and in >> their brain only. So providing this would be a welcome improvement to me. >> >> AI coders often dont manage to implement the idea of the prompt into a >> working system but thats another issue ;) >> >> Anything is welcome that helps prevent the danger of being coerced into >> or accidentally reading sloppy output (i dont care if the low quality is >> caused by some LLM or human) >> >> On Monday, January 26, 2026, Paul Lalonde <[email protected]> >> wrote: >> >>> I use Markdown/Markdeep syntax. The idea was so I can read my .md files >>> and do basic chores with them in-place. >>> The source text is all in body.all, and the renderer interprets that. I >>> keep the selection aligned between the two in the hope I can come to >>> copy/paste with markup semantics holding up. That's full of edge cases, >>> but it's already useful enough that I'm now editing my .md docs for the AI >>> in Edwood itself. >>> >>> This is entirely eyes-off code. There is zero value, near as I can >>> tell, in the code artifact. My specs and debug sessions have the value. >>> This is largely an exercise in how to iterate on the spec to get a >>> repeatable artifact as output, but that's still far from good - I don't >>> have a sufficient way to automate the interaction testing. >>> >>> Paul >>> >>> On Mon, Jan 26, 2026 at 11:15 AM Shawn Rutledge <[email protected]> >>> wrote: >>> >>>> > On Jan 26, 2026, at 07:21, Paul Lalonde <[email protected]> >>>> wrote: >>>> > >>>> > I've wanted rich text in acme (well, Edwood) windows for a long time. >>>> > Claude and enough specification seems entirely capable of doing what >>>> I hadn't been able to find the time to do myself. Not ready for >>>> distribution, and the code itself has negative value, but the suite of >>>> tests and bundle of specs may have enough value to make this reproducible. >>>> >>>> Nice! >>>> >>>> But how do you specify the formatting? >>>> >>> *9fans <https://9fans.topicbox.com/latest>* / 9fans / see discussions >>> <https://9fans.topicbox.com/groups/9fans> + participants >>> <https://9fans.topicbox.com/groups/9fans/members> + delivery options >>> <https://9fans.topicbox.com/groups/9fans/subscription> Permalink >>> <https://9fans.topicbox.com/groups/9fans/T3fa6a271939d2a0b-M163ab075e6ac11871ff07dba> >>> >> *9fans <https://9fans.topicbox.com/latest>* / 9fans / see discussions > <https://9fans.topicbox.com/groups/9fans> + participants > <https://9fans.topicbox.com/groups/9fans/members> + delivery options > <https://9fans.topicbox.com/groups/9fans/subscription> Permalink > <https://9fans.topicbox.com/groups/9fans/T3fa6a271939d2a0b-M8d39f6aca4ed5e25c5c659a0> > ------------------------------------------ 9fans: 9fans Permalink: https://9fans.topicbox.com/groups/9fans/T3fa6a271939d2a0b-M70c40b3e25b970a0499e131f Delivery options: https://9fans.topicbox.com/groups/9fans/subscription
