Morty

20+ yrs software developer (Mobile, Backend, Web, and passionate for AI). I write about my experiences: Coding and Life

[Morty @m0rtitech]

@lucianghinda two files with the same rules always drift imo. keep one real file and let the other point to it

[Morty @m0rtitech]

@RoundtableSpace spec first is the right order imo. unit of work moved from story to spec, agents just made it obvious

[Morty @m0rtitech]

@elonmusk @cradrrat Please fix the transcript as well. Most of the time, wrong words transcribed by the bot need correction.

[Morty @m0rtitech]

@elonmusk @cradrrat @elonmusk The lack of widely used connectors is Grok Bot's biggest weakness now. Otherwise, it's getting better every day.

[Morty @m0rtitech]

@badlogicgames honestly underrated. llm labels the data once, then a tiny model does the boring part fast and cheap

[Morty @m0rtitech]

@theo a transfer impact section right in the PR is the smart part imo. makes the cost visible before merge, not after

[Morty @m0rtitech]

@simonw funny but makes sense. tell it the code is its own and it'll refactor more boldly.. which is great only if the tests are there to catch it

[Morty @m0rtitech]

@PawelHuryn the cost per fix is the real story tbh. fable ~$2.40 per bug vs sol ~$0.13, so 18x more for 14 more bugs. for hard bugs worth it, for a backlog sweep i'd route the easy ones to the cheap model first

[Morty @m0rtitech]

@GergelyOrosz half yes imo. waterfall didnt fail only bc coding was slow, it failed bc u found out the plan was wrong only after users saw it. so plan upfront but in small specs u can throw away, rewriting a spec is cheap now too

[Morty @m0rtitech]

@elonmusk @bot Hey @elonmusk, there's a problem with connectors, especially 1Password. The 'Shared With Grok bot' vault is no longer accessible and defaults to the personal vault, so I have to enter the credentials manually.

[Morty @m0rtitech]

@simonw like it, but i'd put one more thing in the human part above the fold.. what the reviewer should actually check and what the tests prove. the long agent essay is fine hidden, the risky bit shouldnt be

[Morty @m0rtitech]

@RoundtableSpace 3 runs each and the nova ranges overlap, 15-18 vs 12-19, so thats more a tie than a win imo. the cost part worries me more tbh, under $3 a run vs ~$22 and deepseek did most of its progress after 100M tokens .

[Morty @m0rtitech]

@swyx same reason a spec file beats a long chat with an agent tbh. writing it down is where the thinking happens

[Morty @m0rtitech]

@theo a title that changes while the agent works is worse than a few lost pixels imo. 2 line titles sound worth it

[Morty @m0rtitech]

@elonmusk it’s burn tokens like hell. I almost can’t use it efficiently

[Morty @m0rtitech]

@GergelyOrosz @dillon_mulroy @championswimmer dry run first imo. have it print the exact handles before it blocks anyone

[Morty @m0rtitech]

@addyosmani two failed corrections then restart is the underrated part tbh. a third try is just more junk it has to reread

[Morty @m0rtitech]

@RoundtableSpace 413 files for one badge is the whole lesson. keep memory under 200 lines

[Morty @m0rtitech]

@MalcolmLerider prompts arent a lock. delete and drop need a hard gate in code imo

[Morty @m0rtitech]

@MengTo opus on three.js is wild but the hard part is still taste imo