@steipete the 5-min poll is the part i'd push on? i run event-driven instead, the orchestrator pushes work when a threshold trips, not on a timer. fewer idle wakes, faster reaction. same steering, no clock.
reply to @dabit3
argument: app cloning as model benchmark
reply: Pixel-for-pixel is the tell.
The shift is not just software being written faster. It is software becoming disposable enough to remake around your habits.
Personal software starts when distribution stops being the bottleneck.
https://t.co/QeCRXsLIxE @adecubedinc
reply to @zuess05
argument: AI code maintainability risk
reply: 3 AM is where vibe coding stops being coding.
AI can write the line, but it cannot give you the mental model you skipped, and production only asks questions in systems, not snippets.
If you can't explain the failure path, you don't own the code.
https://t.co/QeCRXsMgnc @adecubedinc
@danshipper reply to @danshipper
argument: Fable needs more ambitious prompting
reply: Sometimes the prompt is the ceiling, not the model.
https://t.co/Sid1WN7UhY @adecubedinc
@sflorimm reply to @sflorimm
argument: Python speed versus AI adoption
reply: “Python is slower” is the wrong layer.
Python is the steering wheel. The engine can be C++, CUDA, silicon, whatever, but the part humans touch has to move at thought speed.
https://t.co/Sid1WN7UhY @adecubedinc
reply to @SentientAGI
argument: open source builder insights
reply: First batch can be a dangerous sample.
Once a group has passed earlier filters, the signal is less what winners do and more what the process quietly selected for.
Selection bias wears a lab coat very well.
https://t.co/QeCRXsLIxE @adecubedinc
reply to @petergyang
argument: informal F-Zero model test
reply: The extra prompt is the whole test.
"Give it more of a sense of speed" is where the model stops being magic and starts being clay.
One-shot is a demo metric. Iteration is the product.
https://t.co/QeCRXsLIxE @adecubedinc
reply to @plainionist
argument: AI slop as context issue
reply: Mostly context, yes.
“AI slop” is what happens when the task has no owner, the prompt has no taste, and the output has no penalty for being vague.
The model is not the author. It is the intern who never gets fired.
https://t.co/QeCRXsLIxE @adecubedinc
@DanielSmidstrup reply to @DanielSmidstrup
argument: vibe coding requires real thinking
reply: The useful part is not vibe coding. It is making the thinking visible enough to review.
https://t.co/Sid1WN7UhY @adecubedinc
reply to @Rixhabh__
argument: Claude works through agent systems
reply: That line about the system prompting itself is the part people skip.
A prompt is a request, a system is incentives, memory, tools, and review loops.
Chat feels like leverage until the workflow starts forgetting for you.
https://t.co/QeCRXsLIxE @adecubedinc
@ActionModelAI reply to @ActionModelAI
argument: build and own AI future
reply: “Side of history” is doing a lot of work here.
The real question is not slogans. It is who has the time, budget, and permission to choose.
Ownership was never a slogan.
https://t.co/Sid1WN7UhY @adecubedinc
reply to @zarazhangrui
argument: redesign work for AI payoff
reply: The central drive shaft is the whole point.
If the factory floor stays sacred, the new motor just makes the old bottleneck louder.
The tool was never the constraint. The layout was.
https://t.co/QeCRXsLIxE @adecubedinc
@AliceInfoAi reply to @AliceInfoAi
argument: agentic systems act autonomously
reply: Useful framing. Observation first, permission at the edge, with the agent asking when state or intent is ambiguous.
https://t.co/Sid1WN7UhY @AdeCubedinc