Mastodawn

I prefer intention (setup what I want & meet expectations=no waste & no rejection) over inspection (check if what is offered meets expectations=more waste & more rejection).
Producing code with AI-LLM based agents/tools (scripts/APIs masking as some kind of glorified intelligence) is akin to going into a women’s clothing store and having to try on every item (review all produced code) you want to buy for personal fit(code accuracy/security/performance) as the size indicators don’t mean anything.

Show thread

Dawn Ahukanna

Invest personally where you choose.

“vibe coding is like quick-set cement with superglue, you can only pour and set once”. If and when you need to make any changes, it’s a complete redo, including removing what is already there.

So it’s appropriate for “closed system/laboratory” prototypes, not “open system/real world” production software systems.

#AIEthics #techConsequencesInRealWorld

Show thread

Dawn Ahukanna Mar 22

Related-https://mastodon.social/@dahukanna/116269110801236174
‘This tech’s intelligent, has “hovered” up public Internet & now humans need to tell the tech what to do?
“Tell AI agents how to operate your website or app. … When agent lands on website, it is blindfolded.
It reconstructs entire interface from screenshots & DOM trees, guesses what buttons do, mistakes loading screens for errors, wrong actions on irreversible operations & wastes compute figuring out things you could have just told it.” 🤦🏾‍♀️
- https://infosec.exchange/@Markcarter/116267086205344350

mark carter (@[email protected])

Powerful and simple 🧙 operate.txt — A standard for the agentic web. Tell AI agents how to operate your website or app. We are no longer in the "chat with AI" era. We are in the era where AI agents act on your behalf — clicking buttons, filling forms, completing purchases, navigating workflows. Claude Projects. Persistent memory. Computer use. Multi-agent orchestration. These are production features, not demos. Agents are operating on real websites, apps, and tools right now. And when an agent lands on your website, it is blindfolded. It reconstructs your entire interface from screenshots and DOM trees. It guesses what buttons do. It mistakes loading screens for errors. It takes wrong actions on irreversible operations. It wastes compute figuring out things you could have just told it. Operate.txt is an "operate me" file. It tells agents how your product works, how to navigate it, what actions exist, and what consequences those actions have. It's the missing layer between your interface and the agents your users are already deploying. https://github.com/serdem1/operate.txt #AI #UX

Infosec Exchange

Show thread

Dawn Ahukanna Mar 23

Related - https://social.kernel.org/objects/c67b32ed-cf17-4612-b725-b40b701287bc
It’s “one and done” with superglued code, not “nurture, maintain and mature”.
The productivity is all about the speed of production (utility), not the usefulness or quality or experience (delight).

Jonathan Corbet (@[email protected])

As the number of LLM-generated patches in my inbox increases, I am starting to experience the sort of maintainer stress that has long been predicted. But there's another aspect of this that has re...

Show thread

Dawn Ahukanna Mar 24

“reminds me of stuff that Kate Compton is working on. 🤔 haven't read her thesis yet, maybe she actually cites ‘programming as theory building’ by peter naur, 1985
^__^

https://galaxykate.com/